feat: hot-reload voice models without server restart

Voice enabled/STT model are now DB-backed (admin settings), not env
vars. Added reload_stt_model()/reload_tts_model() that clear singletons
under lock and re-trigger loading. POST /api/admin/voice/reload triggers
both in background tasks. Settings UI polls /api/voice/status every 2.5s
until models are ready, with spinner feedback.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

docker-compose.yml

@@ -15,7 +15,7 @@ services:
     environment:
       DATABASE_URL: "postgresql+asyncpg://${POSTGRES_USER:-fabled}:${POSTGRES_PASSWORD:-fabled}@db:5432/${POSTGRES_DB:-fabledassistant}"
       OLLAMA_URL: "http://ollama:11434"
-      OLLAMA_MODEL: "${OLLAMA_MODEL:-llama3.1}"
+      OLLAMA_MODEL: "${OLLAMA_MODEL:-qwen3:8B}"
       SECRET_KEY: "${SECRET_KEY:-dev-secret-change-me}"
       # Uncomment and set to enable web research and image search via SearXNG:
       # SEARXNG_URL: "http://searxng:8080"