refactor(llm_models): 将用量记录函数改为异步以支持数据库写入

将 `LLMUsageRecorder` 中的 `record_usage_to_database` 和 `LLMRequest` 中的 `_record_usage` 方法从同步转换为异步。

此项重构是为了支持异步数据库操作，避免在记录模型用量时阻塞事件循环，从而提升应用的整体性能和响应能力。

src/llm_models/utils.py

@@ -145,7 +145,7 @@ class LLMUsageRecorder:
     LLM使用情况记录器（SQLAlchemy版本）
     """
 
-    def record_usage_to_database(
+    async def record_usage_to_database(
         self,
         model_info: ModelInfo,
         model_usage: UsageRecord,

src/llm_models/utils_model.py

@@ -891,7 +891,7 @@ class LLMRequest:
             max_tokens=self.model_for_task.max_tokens if max_tokens is None else max_tokens,
         )
 
-        self._record_usage(model_info, response.usage, time.time() - start_time, "/chat/completions")
+        await self._record_usage(model_info, response.usage, time.time() - start_time, "/chat/completions")
 
         if not response.content and not response.tool_calls:
             if raise_when_empty:
@@ -916,14 +916,14 @@ class LLMRequest:
             embedding_input=embedding_input
         )
         
-        self._record_usage(model_info, response.usage, time.time() - start_time, "/embeddings")
+        await self._record_usage(model_info, response.usage, time.time() - start_time, "/embeddings")
         
         if not response.embedding:
             raise RuntimeError("获取embedding失败")
         
         return response.embedding, model_info.name
 
-    def _record_usage(self, model_info: ModelInfo, usage: Optional[UsageRecord], time_cost: float, endpoint: str):
+    async def _record_usage(self, model_info: ModelInfo, usage: Optional[UsageRecord], time_cost: float, endpoint: str):
         """
         记录模型使用情况。