feat(llm): let each model tier run on its own provider (#1440)

- quick_think_provider / deep_think_provider and their backend URLs (with TRADINGAGENTS_* variables) choose a tier's provider; unset, a tier uses llm_provider and backend_url
- a tier on another provider gets that provider's own settings and endpoint, and the run settings and report header name each tier's provider
- the CLI checks each tier provider's model variable and key before the run

cli/selections.py

@@ -68,6 +68,29 @@ def unattended_gaps(flags) -> list[str]:
     return gaps
 
 
+def _check_tier_providers(main_provider: str) -> None:
+    """Check each provider the model tiers use (#1440) before the run.
+
+    A tier on another provider needs its model from its variable, since the
+    model question offers the main provider's models. Each provider a tier uses
+    has its key checked now, prompting for a missing one, rather than at its
+    first call, which for the deep tier comes after everything else has been
+    paid for; a provider no tier uses needs no key.
+    """
+    used = []
+    for tier in ("quick", "deep"):
+        provider = (DEFAULT_CONFIG.get(f"{tier}_think_provider") or main_provider).lower()
+        if provider != main_provider.lower():
+            variable = f"TRADINGAGENTS_{tier.upper()}_THINK_LLM"
+            if not os.environ.get(variable):
+                console.print(f"[red]The {tier} tier runs on {provider}; set {variable} to one of its models.[/red]")
+                raise typer.Exit(code=1)
+        if provider not in used:
+            used.append(provider)
+    for provider in used:
+        ensure_api_key(provider)
+
+
 def _from_flag(parse, value, *args):
     """A flag's value through the same check its prompt applies; a bad one ends the run."""
     try:
@@ -223,8 +246,6 @@ def _prompt_selections(prefs, flags):
         )
         console.print(f"[green]✓ LLM provider from environment:[/green] {selected_llm_provider}")
         console.print(f"[green]✓ Backend URL:[/green] {backend_url}")
-        # Still confirm/persist the API key so the run doesn't fail later.
-        ensure_api_key(selected_llm_provider)
     else:
         console.print(
             create_question_box(
@@ -261,10 +282,8 @@ def _prompt_selections(prefs, flags):
         if selected_llm_provider == "ollama":
             confirm_ollama_endpoint(backend_url)
 
-        # Confirm the provider's API key is present; prompt the user to paste
-        # one and persist it to .env if it's missing, so the analysis run
-        # doesn't fail later at the first API call.
-        ensure_api_key(selected_llm_provider)
+
+    _check_tier_providers(selected_llm_provider)
 
     # Step 7: Thinking agents (skipped when either model is set via environment)
     if os.environ.get("TRADINGAGENTS_QUICK_THINK_LLM") or os.environ.get("TRADINGAGENTS_DEEP_THINK_LLM"):