Phase 17: Multimodal image understanding via analyze_image tool

Dual-model approach (C): Qwen3-8B handles conversation, Qwen2.5-VL-7B
analyzes images on demand via analyze_image LangChain tool.

- services/model/mlx_vision_model.py: MlxVisionModel (mlx-vlm wrapper, lazy load)
- services/agent/tools.py: make_vision_tool(vision_model, image_path)
- agent_service.py: stream_response(image_path=None), dynamic tool binding
  via config["image_path"] — thread-safe per-request rebinding
- container.py: vision_model Singleton provider
- config.py: vision_enabled, vision_model_id, vision_max_tokens
- api.py: image_base64 in ChatRequest, decode to temp file, cleanup after stream

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

services/agent/tools.py

@@ -3,6 +3,17 @@ from datetime import date
 from langchain_core.tools import tool
 
 
+def make_vision_tool(vision_model, image_path: str):
+    """현재 요청에 첨부된 이미지를 분석하는 도구."""
+
+    @tool
+    def analyze_image(prompt: str = "이 이미지를 한국어로 자세히 설명해줘.") -> str:
+        """첨부된 이미지를 분석한다. 이미지 속 음식, 문서, 사람, 사물 등을 파악할 때 사용하세요."""
+        return vision_model.analyze(image_path, prompt)
+
+    return analyze_image
+
+
 @tool
 def get_current_date() -> str:
     """오늘 날짜를 반환합니다. 나이 계산, 날짜 비교 등 현재 날짜가 필요할 때 반드시 먼저 호출하세요."""