fix: 修复本地llm服务不可用问题 + 统一模型缓存目录位置

- 修复 config.py 添加 LOCAL_MODEL_NAME 配置项
- 修复 chat_services.py 本地模型检测时API路径重复问题(/v1/models -> /models)
- 更新 .gitignore，移除模型目录跟踪
- 统一模型缓存到 docker/models/fastembed_cache，避免重复
- 更新 Dockerfile，正确复制预下载的BM25模型缓存

backend/app/model_services/chat_services.py

@@ -27,7 +27,8 @@ from app.config import (
     VLLM_BASE_URL,
     LLM_API_KEY,
     ZHIPUAI_API_KEY,
-    DEEPSEEK_API_KEY
+    DEEPSEEK_API_KEY,
+    LOCAL_MODEL_NAME
 )
 
 logger = logging.getLogger(__name__)
@@ -38,9 +39,9 @@ class LocalVLLMChatProvider(BaseServiceProvider[BaseChatModel]):
     本地 VLLM 生成式大模型服务提供者
     """
     
-    def __init__(self, model: str = "gemma-4-E4B-it"):
+    def __init__(self, model: str = None):
         super().__init__("local_vllm_chat")
-        self._model = model
+        self._model = model or LOCAL_MODEL_NAME
     
     def is_available(self) -> bool:
         """
@@ -80,7 +81,7 @@ class LocalVLLMChatProvider(BaseServiceProvider[BaseChatModel]):
                 headers["Authorization"] = f"Bearer {LLM_API_KEY}"
             
             try:
-                response = client.get("/v1/models", headers=headers)
+                response = client.get("/models", headers=headers)
                 if response.status_code == 200:
                     logger.info(f"本地 VLLM 服务可用: {self._model}")
                     return True