feat: AI 서비스 MLX 듀얼 백엔드 및 모델 최적화

- MLX(맥미니 27B) 우선 → Ollama(조립컴 9B) fallback 구조
- pydantic-settings 기반 config 전환
- health check에 MLX 상태 추가
- 텍스트 모델 qwen3:8b → qwen3.5:9b-q8_0 변경

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

ai-service/routers/embeddings.py

@@ -1,4 +1,4 @@
-from fastapi import APIRouter, BackgroundTasks, Query
+from fastapi import APIRouter, BackgroundTasks, HTTPException, Query
 from pydantic import BaseModel
 from services.embedding_service import (
     sync_all_issues,
@@ -53,7 +53,7 @@ async def get_similar(issue_id: int, n_results: int = Query(default=5, le=20)):
         results = await search_similar_by_id(issue_id, n_results)
         return {"available": True, "results": results, "query_issue_id": issue_id}
     except Exception as e:
-        return {"available": False, "results": [], "error": str(e)}
+        raise HTTPException(status_code=500, detail="AI 서비스 처리 중 오류가 발생했습니다")
 
 
 @router.post("/similar/search")
@@ -69,7 +69,7 @@ async def search_similar(req: SearchRequest):
         )
         return {"available": True, "results": results}
     except Exception as e:
-        return {"available": False, "results": [], "error": str(e)}
+        raise HTTPException(status_code=500, detail="AI 서비스 처리 중 오류가 발생했습니다")
 
 
 @router.get("/embeddings/stats")