feat: structured-output Trader and Research Manager (#434, finishes the trio)

Extends the canonical structured-output pattern from the Portfolio Manager
to the other two decision-making agents.  Each of the three agents now
returns a typed Pydantic instance via llm.with_structured_output() in a
single primary call, and a render helper turns the result into the same
markdown shape downstream agents and saved reports already consume.

- ResearchPlan: 5-tier recommendation, conversational rationale, concrete
  strategic actions for the trader.
- TraderProposal: 3-tier action (transaction direction is naturally Buy /
  Hold / Sell — position sizing happens later at the Portfolio Manager),
  reasoning, and optional entry_price / stop_loss / position_sizing.
  Rendered output preserves the trailing "FINAL TRANSACTION PROPOSAL:
  **BUY/HOLD/SELL**" line for backward compatibility with the analyst
  stop-signal text.
- PortfolioDecision: 5-tier rating, executive summary, investment thesis,
  optional price_target / time_horizon (unchanged).

The shared try-structured-then-fallback pattern is extracted into
tradingagents/agents/utils/structured.py (bind_structured +
invoke_structured_or_freetext) so all three agents go through the same
code path and log the same warning when a provider lacks structured
output and the agent falls back to free-text generation.

Net effect for users: every saved markdown report (research/manager.md,
trading/trader.md, portfolio/decision.md) now has consistent section
headers across runs and providers, easier to scan.

Net effect for the runtime: the rating extraction round-trip is gone —
the rating comes from the structured response itself, not a second
LLM call. SignalProcessor was already simplified to a heuristic adapter
in the previous commit.

11 new tests in tests/test_structured_agents.py cover the Trader and
Research Manager render functions, structured-output happy paths, and
free-text fallback. Full suite: 88 tests pass in ~2s without API keys.

tests/test_structured_agents.py

@@ -0,0 +1,232 @@
+"""Tests for structured-output agents (Trader and Research Manager).
+
+The Portfolio Manager has its own coverage in tests/test_memory_log.py
+(which exercises the full memory-log → PM injection cycle).  This file
+covers the parallel schemas, render functions, and graceful-fallback
+behavior we added for the Trader and Research Manager so all three
+decision-making agents share the same shape.
+"""
+
+from unittest.mock import MagicMock
+
+import pytest
+
+from tradingagents.agents.managers.research_manager import create_research_manager
+from tradingagents.agents.schemas import (
+    PortfolioRating,
+    ResearchPlan,
+    TraderAction,
+    TraderProposal,
+    render_research_plan,
+    render_trader_proposal,
+)
+from tradingagents.agents.trader.trader import create_trader
+
+
+# ---------------------------------------------------------------------------
+# Render functions
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.unit
+class TestRenderTraderProposal:
+    def test_minimal_required_fields(self):
+        p = TraderProposal(action=TraderAction.HOLD, reasoning="Balanced setup; no edge.")
+        md = render_trader_proposal(p)
+        assert "**Action**: Hold" in md
+        assert "**Reasoning**: Balanced setup; no edge." in md
+        # The trailing FINAL TRANSACTION PROPOSAL line is preserved for the
+        # analyst stop-signal text and any external code that greps for it.
+        assert "FINAL TRANSACTION PROPOSAL: **HOLD**" in md
+
+    def test_optional_fields_included_when_present(self):
+        p = TraderProposal(
+            action=TraderAction.BUY,
+            reasoning="Strong technicals + fundamentals.",
+            entry_price=189.5,
+            stop_loss=178.0,
+            position_sizing="6% of portfolio",
+        )
+        md = render_trader_proposal(p)
+        assert "**Action**: Buy" in md
+        assert "**Entry Price**: 189.5" in md
+        assert "**Stop Loss**: 178.0" in md
+        assert "**Position Sizing**: 6% of portfolio" in md
+        assert "FINAL TRANSACTION PROPOSAL: **BUY**" in md
+
+    def test_optional_fields_omitted_when_absent(self):
+        p = TraderProposal(action=TraderAction.SELL, reasoning="Guidance cut.")
+        md = render_trader_proposal(p)
+        assert "Entry Price" not in md
+        assert "Stop Loss" not in md
+        assert "Position Sizing" not in md
+        assert "FINAL TRANSACTION PROPOSAL: **SELL**" in md
+
+
+@pytest.mark.unit
+class TestRenderResearchPlan:
+    def test_required_fields(self):
+        p = ResearchPlan(
+            recommendation=PortfolioRating.OVERWEIGHT,
+            rationale="Bull case carried; tailwinds intact.",
+            strategic_actions="Build position over two weeks; cap at 5%.",
+        )
+        md = render_research_plan(p)
+        assert "**Recommendation**: Overweight" in md
+        assert "**Rationale**: Bull case carried" in md
+        assert "**Strategic Actions**: Build position" in md
+
+    def test_all_5_tier_ratings_render(self):
+        for rating in PortfolioRating:
+            p = ResearchPlan(
+                recommendation=rating,
+                rationale="r",
+                strategic_actions="s",
+            )
+            md = render_research_plan(p)
+            assert f"**Recommendation**: {rating.value}" in md
+
+
+# ---------------------------------------------------------------------------
+# Trader agent: structured happy path + fallback
+# ---------------------------------------------------------------------------
+
+
+def _make_trader_state():
+    return {
+        "company_of_interest": "NVDA",
+        "investment_plan": "**Recommendation**: Buy\n**Rationale**: ...\n**Strategic Actions**: ...",
+    }
+
+
+def _structured_trader_llm(captured: dict, proposal: TraderProposal | None = None):
+    """Build a MagicMock LLM whose with_structured_output binding captures the
+    prompt and returns a real TraderProposal so render_trader_proposal works.
+    """
+    if proposal is None:
+        proposal = TraderProposal(
+            action=TraderAction.BUY,
+            reasoning="Strong setup.",
+        )
+    structured = MagicMock()
+    structured.invoke.side_effect = lambda prompt: (
+        captured.__setitem__("prompt", prompt) or proposal
+    )
+    llm = MagicMock()
+    llm.with_structured_output.return_value = structured
+    return llm
+
+
+@pytest.mark.unit
+class TestTraderAgent:
+    def test_structured_path_produces_rendered_markdown(self):
+        captured = {}
+        proposal = TraderProposal(
+            action=TraderAction.BUY,
+            reasoning="AI capex cycle intact; institutional flows constructive.",
+            entry_price=189.5,
+            stop_loss=178.0,
+            position_sizing="6% of portfolio",
+        )
+        llm = _structured_trader_llm(captured, proposal)
+        trader = create_trader(llm)
+        result = trader(_make_trader_state())
+        plan = result["trader_investment_plan"]
+        assert "**Action**: Buy" in plan
+        assert "**Entry Price**: 189.5" in plan
+        assert "FINAL TRANSACTION PROPOSAL: **BUY**" in plan
+        # The same rendered markdown is also added to messages for downstream agents.
+        assert plan in result["messages"][0].content
+
+    def test_prompt_includes_investment_plan(self):
+        captured = {}
+        llm = _structured_trader_llm(captured)
+        trader = create_trader(llm)
+        trader(_make_trader_state())
+        # The investment plan is in the user message of the captured prompt.
+        prompt = captured["prompt"]
+        assert any("Proposed Investment Plan" in m["content"] for m in prompt)
+
+    def test_falls_back_to_freetext_when_structured_unavailable(self):
+        plain_response = (
+            "**Action**: Sell\n\nGuidance cut hits margins.\n\n"
+            "FINAL TRANSACTION PROPOSAL: **SELL**"
+        )
+        llm = MagicMock()
+        llm.with_structured_output.side_effect = NotImplementedError("provider unsupported")
+        llm.invoke.return_value = MagicMock(content=plain_response)
+        trader = create_trader(llm)
+        result = trader(_make_trader_state())
+        assert result["trader_investment_plan"] == plain_response
+
+
+# ---------------------------------------------------------------------------
+# Research Manager agent: structured happy path + fallback
+# ---------------------------------------------------------------------------
+
+
+def _make_rm_state():
+    return {
+        "company_of_interest": "NVDA",
+        "investment_debate_state": {
+            "history": "Bull and bear arguments here.",
+            "bull_history": "Bull says...",
+            "bear_history": "Bear says...",
+            "current_response": "",
+            "judge_decision": "",
+            "count": 1,
+        },
+    }
+
+
+def _structured_rm_llm(captured: dict, plan: ResearchPlan | None = None):
+    if plan is None:
+        plan = ResearchPlan(
+            recommendation=PortfolioRating.HOLD,
+            rationale="Balanced view across both sides.",
+            strategic_actions="Hold current position; reassess after earnings.",
+        )
+    structured = MagicMock()
+    structured.invoke.side_effect = lambda prompt: (
+        captured.__setitem__("prompt", prompt) or plan
+    )
+    llm = MagicMock()
+    llm.with_structured_output.return_value = structured
+    return llm
+
+
+@pytest.mark.unit
+class TestResearchManagerAgent:
+    def test_structured_path_produces_rendered_markdown(self):
+        captured = {}
+        plan = ResearchPlan(
+            recommendation=PortfolioRating.OVERWEIGHT,
+            rationale="Bull case is stronger; AI tailwind intact.",
+            strategic_actions="Build position gradually over two weeks.",
+        )
+        llm = _structured_rm_llm(captured, plan)
+        rm = create_research_manager(llm)
+        result = rm(_make_rm_state())
+        ip = result["investment_plan"]
+        assert "**Recommendation**: Overweight" in ip
+        assert "**Rationale**: Bull case" in ip
+        assert "**Strategic Actions**: Build position" in ip
+
+    def test_prompt_uses_5_tier_rating_scale(self):
+        """The RM prompt must list all five tiers so the schema enum matches user expectations."""
+        captured = {}
+        llm = _structured_rm_llm(captured)
+        rm = create_research_manager(llm)
+        rm(_make_rm_state())
+        prompt = captured["prompt"]
+        for tier in ("Buy", "Overweight", "Hold", "Underweight", "Sell"):
+            assert f"**{tier}**" in prompt, f"missing {tier} in prompt"
+
+    def test_falls_back_to_freetext_when_structured_unavailable(self):
+        plain_response = "**Recommendation**: Sell\n\n**Rationale**: ...\n\n**Strategic Actions**: ..."
+        llm = MagicMock()
+        llm.with_structured_output.side_effect = NotImplementedError("provider unsupported")
+        llm.invoke.return_value = MagicMock(content=plain_response)
+        rm = create_research_manager(llm)
+        result = rm(_make_rm_state())
+        assert result["investment_plan"] == plain_response