Refactor: Zentrales Logging + robuste Token-Erfassung (Phase 2)

- app.py: logging.basicConfig (MC_LOG_LEVEL, INFO default) als eine
  Konfiguration für alle Module.
- Neuer services/gateway_stream.py: SSE-/Non-Stream-usage-Parsing aus dem
  gateway_proxy-Router extrahiert; robuster Zeilenparser mit Debug-Logging
  statt verschluckter Exceptions. Router ist jetzt dünn.
- token_stats.py: In-Memory-Cache + gedrosseltes Flushen (5s) + atexit-Flush
  statt Write-pro-Request; atomarer Write (.tmp -> replace); thread-safe.
- agent.py/discover.py: stille `except Exception: pass` durch gezieltes
  log.debug/warning ersetzt; ungenutzten yaml-Import entfernt.

Verifiziert: Stream-Parsing (Summen + per-Modell), malformed-Chunk übersteht,
flush schreibt; app importiert sauber.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

backend/routers/gateway_proxy.py

@@ -1,11 +1,10 @@
-import json
 import httpx
 from fastapi import APIRouter, Request
 from fastapi.responses import JSONResponse, StreamingResponse
 
 from config import LLAMA_SWAP_URL
+from services.gateway_stream import record_stream_chunk, record_usage
 from services.router_logic import FAST, FAST_NO_THINK, choose_model
-from services.token_stats import increment_tokens
 
 router = APIRouter(prefix="/v1")
 
@@ -37,41 +36,14 @@ async def _proxy(path: str, request: Request):
             async with httpx.AsyncClient(timeout=None) as c:
                 async with c.stream("POST", url, json=body) as r:
                     async for chunk in r.aiter_raw():
-                        try:
-                            chunk_str = chunk.decode("utf-8", errors="ignore")
-                            if '"usage":' in chunk_str:
-                                for line in chunk_str.splitlines():
-                                    if line.startswith("data:"):
-                                        data_str = line[5:].strip()
-                                        if data_str == "[DONE]":
-                                            continue
-                                        try:
-                                            data_json = json.loads(data_str)
-                                            usage = data_json.get("usage")
-                                            if usage:
-                                                prompt = usage.get("prompt_tokens", 0)
-                                                completion = usage.get("completion_tokens", 0)
-                                                if prompt or completion:
-                                                    increment_tokens(prompt, completion, model=alias)
-                                        except Exception:
-                                            pass
-                        except Exception:
-                            pass
+                        record_stream_chunk(chunk, alias)
                         yield chunk
         return StreamingResponse(gen(), media_type="text/event-stream", headers=routed)
 
     async with httpx.AsyncClient(timeout=600) as c:
         r = await c.post(url, json=body)
         resp_json = r.json()
-        try:
-            usage = resp_json.get("usage")
-            if usage:
-                prompt = usage.get("prompt_tokens", 0)
-                completion = usage.get("completion_tokens", 0)
-                if prompt or completion:
-                    increment_tokens(prompt, completion, model=alias)
-        except Exception:
-            pass
+        record_usage(resp_json.get("usage") if isinstance(resp_json, dict) else None, alias)
         return JSONResponse(resp_json, status_code=r.status_code, headers=routed)