refactor: replace print() with structured logging across backend

Replace all print()/stderr logging with Python logging module using
logger = logging.getLogger(__name__) pattern for consistent log levels
and formatting. Extract score conversion to shared score_utils module.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

backend/src/llm/siliconflow.py

@@ -1,6 +1,7 @@
 """
 大模型API集成 — 支持 SiliconFlow 和 DeepSeek 官方 + 视觉模型
 """
+import logging
 import os
 import base64
 from typing import List, Dict, Any, Optional, AsyncGenerator, Tuple
@@ -12,6 +13,8 @@ import openai
 
 from ..core.config import get_settings
 
+logger = logging.getLogger(__name__)
+
 settings = get_settings()
 
 # 图片描述提示词
@@ -56,7 +59,7 @@ class SiliconFlowLLM:
         api_key, base_url, resolved_model = _resolve_provider(raw_model)
 
         self.model_name = resolved_model
-        print(f"[LLM] 模型: {resolved_model}, API: {base_url}")
+        logger.info(f"模型: {resolved_model}, API: {base_url}")
 
         self.llm = ChatOpenAI(
             model=resolved_model,
@@ -210,7 +213,7 @@ class SiliconFlowLLM:
                 return response.choices[0].message.content or ""
 
             except Exception as e:
-                print(f"[VLM] 描述失败 attempt={attempt+1}: {e}")
+                logger.warning(f"描述失败 attempt={attempt+1}: {e}")
                 if attempt < max_retries - 1:
                     await asyncio.sleep(2 ** attempt)