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/config.py

@@ -1,24 +1,30 @@
-import os
+from pydantic_settings import BaseSettings
 
 
-class Settings:
-    OLLAMA_BASE_URL: str = os.getenv("OLLAMA_BASE_URL", "http://100.111.160.84:11434")
-    OLLAMA_TEXT_MODEL: str = os.getenv("OLLAMA_TEXT_MODEL", "qwen2.5:14b-instruct-q4_K_M")
-    OLLAMA_EMBED_MODEL: str = os.getenv("OLLAMA_EMBED_MODEL", "bge-m3")
-    OLLAMA_TIMEOUT: int = int(os.getenv("OLLAMA_TIMEOUT", "120"))
+class Settings(BaseSettings):
+    OLLAMA_BASE_URL: str = "http://100.111.160.84:11434"
+    OLLAMA_TEXT_MODEL: str = "qwen3:8b"
+    OLLAMA_EMBED_MODEL: str = "bge-m3"
+    OLLAMA_TIMEOUT: int = 120
 
-    DB_HOST: str = os.getenv("DB_HOST", "mariadb")
-    DB_PORT: int = int(os.getenv("DB_PORT", "3306"))
-    DB_USER: str = os.getenv("DB_USER", "hyungi_user")
-    DB_PASSWORD: str = os.getenv("DB_PASSWORD", "")
-    DB_NAME: str = os.getenv("DB_NAME", "hyungi")
+    MLX_BASE_URL: str = "https://llm.hyungi.net"
+    MLX_TEXT_MODEL: str = "/Users/hyungi/mlx-models/Qwen3.5-27B-4bit"
 
-    SECRET_KEY: str = os.getenv("SECRET_KEY", "")
+    DB_HOST: str = "mariadb"
+    DB_PORT: int = 3306
+    DB_USER: str = "hyungi_user"
+    DB_PASSWORD: str = ""
+    DB_NAME: str = "hyungi"
+
+    SECRET_KEY: str = ""
     ALGORITHM: str = "HS256"
 
-    SYSTEM1_API_URL: str = os.getenv("SYSTEM1_API_URL", "http://system1-api:3005")
-    CHROMA_PERSIST_DIR: str = os.getenv("CHROMA_PERSIST_DIR", "/app/data/chroma")
-    METADATA_DB_PATH: str = os.getenv("METADATA_DB_PATH", "/app/data/metadata.db")
+    SYSTEM1_API_URL: str = "http://system1-api:3005"
+    CHROMA_PERSIST_DIR: str = "/app/data/chroma"
+    METADATA_DB_PATH: str = "/app/data/metadata.db"
+
+    class Config:
+        env_file = ".env"
 
 
 settings = Settings()