feat: DeepSeek V4 thinking-mode round-trip via DeepSeekChatOpenAI subclass

Resolves #599: thinking-mode models require reasoning_content to be
echoed back across turns; multi-turn agent runs failed with HTTP 400.

The fix isolates DeepSeek's quirks (reasoning_content round-trip and
the deepseek-reasoner no-tool_choice limitation) into a subclass so
the general OpenAI-compatible client stays untouched. Adds DeepSeek
V4 Pro/Flash to the catalog. 9 new tests; rationale documented in
the class docstrings.

Design adapted from #600; #611 closed in favour of this approach.

tests/test_deepseek_reasoning.py

@@ -0,0 +1,169 @@
+"""Tests for DeepSeekChatOpenAI thinking-mode behaviour.
+
+Two pieces verified:
+
+1. ``reasoning_content`` is captured on receive into the AIMessage's
+   ``additional_kwargs`` and re-attached on send so DeepSeek's API
+   sees the same value across turns.
+2. ``with_structured_output`` raises NotImplementedError for
+   ``deepseek-reasoner`` so the agent factories' free-text fallback
+   handles the request instead of failing at runtime.
+"""
+
+import os
+
+import pytest
+from langchain_core.messages import AIMessage, HumanMessage
+from langchain_core.prompt_values import ChatPromptValue
+
+from tradingagents.llm_clients.openai_client import (
+    DeepSeekChatOpenAI,
+    NormalizedChatOpenAI,
+    _input_to_messages,
+)
+
+
+# ---------------------------------------------------------------------------
+# _input_to_messages — the helper that handles list / ChatPromptValue / other
+# (Gemini bot review note: non-list inputs must also work)
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.unit
+class TestInputToMessages:
+    def test_list_input_returned_as_is(self):
+        msgs = [HumanMessage(content="hi")]
+        assert _input_to_messages(msgs) is msgs
+
+    def test_chat_prompt_value_unwrapped(self):
+        msgs = [HumanMessage(content="hi")]
+        prompt_value = ChatPromptValue(messages=msgs)
+        assert _input_to_messages(prompt_value) == msgs
+
+    def test_string_input_yields_empty_list(self):
+        # A bare string isn't a message-bearing input; the caller's normal
+        # langchain conversion happens upstream of _get_request_payload.
+        assert _input_to_messages("hello") == []
+
+
+# ---------------------------------------------------------------------------
+# Reasoning content propagation across turns
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.unit
+class TestDeepSeekReasoningContent:
+    def _client(self):
+        os.environ.setdefault("DEEPSEEK_API_KEY", "placeholder")
+        return DeepSeekChatOpenAI(
+            model="deepseek-v4-flash",
+            api_key="placeholder",
+            base_url="https://api.deepseek.com",
+        )
+
+    def test_capture_on_receive(self):
+        """When the response carries reasoning_content, it lands on the
+        AIMessage's additional_kwargs so the next turn can echo it back."""
+        client = self._client()
+        result = client._create_chat_result(
+            {
+                "model": "deepseek-v4-flash",
+                "choices": [
+                    {
+                        "index": 0,
+                        "message": {
+                            "role": "assistant",
+                            "content": "Plan: buy NVDA.",
+                            "reasoning_content": "Step 1: trend is up. Step 2: ...",
+                        },
+                        "finish_reason": "stop",
+                    }
+                ],
+                "usage": {"prompt_tokens": 10, "completion_tokens": 5, "total_tokens": 15},
+            }
+        )
+        ai = result.generations[0].message
+        assert ai.additional_kwargs["reasoning_content"] == "Step 1: trend is up. Step 2: ..."
+
+    def test_propagate_on_send(self):
+        """When an outgoing AIMessage carries reasoning_content, the request
+        payload echoes it on the corresponding message dict."""
+        client = self._client()
+        prior = AIMessage(
+            content="Plan",
+            additional_kwargs={"reasoning_content": "weighed bull case"},
+        )
+        new_user = HumanMessage(content="Refine.")
+        payload = client._get_request_payload([prior, new_user])
+        # Find the assistant message in the payload
+        assistant_dicts = [m for m in payload["messages"] if m.get("role") == "assistant"]
+        assert assistant_dicts, "assistant message missing from outgoing payload"
+        assert assistant_dicts[0]["reasoning_content"] == "weighed bull case"
+
+    def test_propagate_through_chat_prompt_value(self):
+        """Gemini bot review note: non-list inputs (ChatPromptValue) must
+        also propagate reasoning_content."""
+        client = self._client()
+        prior = AIMessage(
+            content="Plan",
+            additional_kwargs={"reasoning_content": "weighed bull case"},
+        )
+        prompt_value = ChatPromptValue(messages=[prior, HumanMessage(content="Refine.")])
+        payload = client._get_request_payload(prompt_value)
+        assistant_dicts = [m for m in payload["messages"] if m.get("role") == "assistant"]
+        assert assistant_dicts[0]["reasoning_content"] == "weighed bull case"
+
+
+# ---------------------------------------------------------------------------
+# deepseek-reasoner: structured output unavailable, falls through to free-text
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.unit
+class TestDeepSeekReasonerStructuredOutput:
+    def test_with_structured_output_raises_for_reasoner(self):
+        client = DeepSeekChatOpenAI(
+            model="deepseek-reasoner",
+            api_key="placeholder",
+            base_url="https://api.deepseek.com",
+        )
+        from pydantic import BaseModel
+
+        class _Sample(BaseModel):
+            answer: str
+
+        with pytest.raises(NotImplementedError):
+            client.with_structured_output(_Sample)
+
+    def test_with_structured_output_works_for_v4(self):
+        """V4 models (non-reasoner) accept tool_choice; structured output works."""
+        client = DeepSeekChatOpenAI(
+            model="deepseek-v4-flash",
+            api_key="placeholder",
+            base_url="https://api.deepseek.com",
+        )
+        from pydantic import BaseModel
+
+        class _Sample(BaseModel):
+            answer: str
+
+        # Should return a Runnable, not raise. (The actual API call would
+        # require a real key; we only assert binding succeeds.)
+        wrapped = client.with_structured_output(_Sample)
+        assert wrapped is not None
+
+
+# ---------------------------------------------------------------------------
+# Base class isolation: NormalizedChatOpenAI does NOT have DeepSeek behaviour
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.unit
+class TestBaseClassIsolation:
+    def test_normalized_does_not_propagate_reasoning_content(self):
+        """The general-purpose NormalizedChatOpenAI must not carry
+        DeepSeek-specific behaviour. Only the subclass does."""
+        assert not hasattr(NormalizedChatOpenAI, "_get_request_payload") or (
+            NormalizedChatOpenAI._get_request_payload
+            is NormalizedChatOpenAI.__bases__[0]._get_request_payload
+        )