feat(memory): automatische Erinnerungs-Extraktion aus Gespraechen

- app/core/memory_extractor.py: LLM destilliert nach je N Turns dauerhafte
  Fakten/Vorlieben aus dem Verlauf, dedupliziert gegen vorhandene Erinnerungen
  und legt sie ab - best-effort, nicht-blockierend (Hintergrund-Task), eigener
  Extraktions-Prompt (JSON, Reasoning aus), Cap-Begrenzung
- Trigger in /api/chat und /ws/voice nach dem Persistieren des Turns
- Konfig: MEMORY_EXTRACTION_ENABLED/_EVERY_N_TURNS/_MAX/_PROVIDER
- Tests: Extraktion, Dedup, kaputtes JSON, Cap, leeres Gespraech, Scheduling
- Doku: README + Architektur-Roadmap (Punkt 3 erledigt)

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

app/api/chat.py

@@ -13,6 +13,7 @@ from app.dependencies import (
     resolve_output_endpoint,
     get_store,
 )
+from app.core.memory_extractor import maybe_schedule_extraction
 from app.quota import enforce_quota, record_usage, QuotaExceededError
 from app.safety.emergency import handle_emergency
 from app.schemas import ChatRequest
@@ -106,6 +107,7 @@ async def chat(
     if session_id:
         store.append_message(session_id, user.id, "user", payload.text)
         store.append_message(session_id, user.id, "assistant", trace.semantic_response)
+        maybe_schedule_extraction(store, user.id, session_id)
 
     if debug:
         return JSONResponse(

app/api/ws.py

@@ -29,6 +29,7 @@ from app.dependencies import (
 )
 from app.store import SessionOwnershipError
 from app.audio.vad import EnergyVAD
+from app.core.memory_extractor import maybe_schedule_extraction
 from app.quota import enforce_quota, record_usage, QuotaExceededError
 from app.safety.emergency import handle_emergency
 
@@ -141,6 +142,7 @@ async def _run_turn(websocket, store, user, session_id, route, orchestrator, out
     if session_id:
         store.append_message(session_id, user.id, "user", text)
         store.append_message(session_id, user.id, "assistant", trace.semantic_response)
+        maybe_schedule_extraction(store, user.id, session_id)
 
     await websocket.send_json(
         {"type": "semantic", "text": trace.semantic_response, "spoken": trace.spoken_response}