AI providers: correlation decision and transcript correction move to OpenAI gpt-4o-mini

One funded account (OpenAI) now covers extraction, embeddings, summaries,
the ambiguous-scene correlation call and transcript correction; Groq stays
optional for STT. New app/internal/llm.py picks the provider from the model
name, so setting either model back to a gemini-* ID needs no code change.
Usage accounting for replay runs now covers both providers.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

drb-c2-core/tests/test_llm_provider.py

@@ -0,0 +1,48 @@
+"""app/internal/llm.py: provider follows the model name (2026-11-08 consolidation)."""
+from types import SimpleNamespace
+from unittest.mock import patch
+
+from app.config import settings
+from app.internal import llm
+
+
+def test_provider_follows_model_name():
+    assert llm.provider_for("gpt-4o-mini") == "openai"
+    assert llm.provider_for("gemini-3.6-flash") == "gemini"
+
+
+def test_available_checks_the_right_key():
+    with patch.object(settings, "openai_api_key", "k"), patch.object(settings, "gemini_api_key", None):
+        assert llm.available("gpt-4o-mini") and not llm.available("gemini-3.6-flash")
+
+
+def test_defaults_are_openai():
+    assert llm.provider_for(settings.corr_cheap_model) == "openai"
+    assert llm.provider_for(settings.transcript_correction_model) == "openai"
+
+
+def test_openai_path_parses_json_and_records_usage():
+    resp = SimpleNamespace(
+        choices=[SimpleNamespace(message=SimpleNamespace(content='{"action": "new"}'))],
+        usage=SimpleNamespace(prompt_tokens=120, completion_tokens=9),
+    )
+    calls = {}
+
+    class FakeClient:
+        def __init__(self):
+            self.chat = SimpleNamespace(completions=SimpleNamespace(create=self.create))
+
+        def create(self, **kw):
+            calls.update(kw)
+            return resp
+
+    sink = {}
+    tok = llm.collect_usage(sink)
+    try:
+        with patch("app.internal.ai_clients.openai_client", FakeClient):
+            out = llm.generate_json("gpt-4o-mini", "Return JSON", purpose="correlation")
+    finally:
+        llm.reset_usage(tok)
+    assert out == {"action": "new"}
+    assert calls["response_format"] == {"type": "json_object"}
+    assert sink["correlation:gpt-4o-mini"] == {"calls": 1, "in": 120, "out": 9, "thinking": 0}