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/app/internal/gemini.py

@@ -16,7 +16,6 @@ what it actually spent instead of an estimate.
 """
 import json
 import threading
-from contextvars import ContextVar
 from typing import Optional
 
 from app.config import settings
@@ -27,18 +26,6 @@ _client_lock = threading.Lock()
 # Models that rejected a thinking level: retried without one from then on.
 _no_thinking_level: set[str] = set()
 
-_usage_sink: ContextVar[Optional[dict]] = ContextVar("drb_gemini_usage", default=None)
-
-
-def collect_usage(sink: Optional[dict]):
-    """Route token counts for the current context into `sink` (a replay run). Returns a reset token."""
-    return _usage_sink.set(sink)
-
-
-def reset_usage(token) -> None:
-    _usage_sink.reset(token)
-
-
 def _get_client():
     global _client
     with _client_lock:
@@ -71,14 +58,8 @@ def _record(purpose: str, model: str, usage) -> None:
     prompt = getattr(usage, "prompt_token_count", None) or 0
     output = getattr(usage, "candidates_token_count", None) or 0
     thoughts = getattr(usage, "thoughts_token_count", None) or 0
-    logger.info(f"gemini usage {purpose} {model}: in={prompt} out={output} thinking={thoughts}")
-    sink = _usage_sink.get()
-    if sink is not None:
-        row = sink.setdefault(f"{purpose}:{model}", {"calls": 0, "in": 0, "out": 0, "thinking": 0})
-        row["calls"] += 1
-        row["in"] += prompt
-        row["out"] += output
-        row["thinking"] += thoughts
+    from app.internal.llm import record_usage
+    record_usage(purpose, model, prompt, output, thoughts)
 
 
 def generate_json(model: str, prompt: str, *, purpose: str,