feat: Enhance emoji and image management with asynchronous background processing

- Added support for scheduling background tasks to build emoji and image descriptions when not found in cache.
- Improved error handling and logging for emoji and image processing.
- Updated `SessionMessage` processing to allow for optional heavy media analysis and voice transcription.
- Refactored logging messages for better clarity and consistency across various modules.
- Introduced a new function to build outbound log previews for messages, enhancing logging capabilities.

src/llm_models/model_client/gemini_client.py

@@ -627,7 +627,7 @@ class GeminiClient(AdapterClient[AsyncIterator[GenerateContentResponse], Generat
             try:
                 thinking_budget = int(extra_params["thinking_budget"])
             except (TypeError, ValueError):
-                logger.warning("无效的 thinking_budget=%s，已回退为自动模式", extra_params["thinking_budget"])
+                logger.warning(f"无效的 thinking_budget={extra_params['thinking_budget']}，已回退为自动模式")
 
         limits: Dict[str, int | bool] | None = None
         if model_id in THINKING_BUDGET_LIMITS:
@@ -646,21 +646,21 @@ class GeminiClient(AdapterClient[AsyncIterator[GenerateContentResponse], Generat
                 return THINKING_BUDGET_DISABLED
             if limits:
                 minimum_value = int(limits["min"])
-                logger.warning("模型 %s 不支持禁用思考预算，已回退为最小值 %s", model_id, minimum_value)
+                logger.warning(f"模型 {model_id} 不支持禁用思考预算，已回退为最小值 {minimum_value}")
                 return minimum_value
             return THINKING_BUDGET_AUTO
 
         if limits is None:
-            logger.warning("模型 %s 未配置思考预算范围，已回退为自动模式", model_id)
+            logger.warning(f"模型 {model_id} 未配置思考预算范围，已回退为自动模式")
             return THINKING_BUDGET_AUTO
 
         minimum_value = int(limits["min"])
         maximum_value = int(limits["max"])
         if thinking_budget < minimum_value:
-            logger.warning("模型 %s 的 thinking_budget=%s 过小，已调整为 %s", model_id, thinking_budget, minimum_value)
+            logger.warning(f"模型 {model_id} 的 thinking_budget={thinking_budget} 过小，已调整为 {minimum_value}")
             return minimum_value
         if thinking_budget > maximum_value:
-            logger.warning("模型 %s 的 thinking_budget=%s 过大，已调整为 %s", model_id, thinking_budget, maximum_value)
+            logger.warning(f"模型 {model_id} 的 thinking_budget={thinking_budget} 过大，已调整为 {maximum_value}")
             return maximum_value
         return thinking_budget
 

src/llm_models/model_client/openai_client.py

@@ -103,7 +103,7 @@ def _normalize_reasoning_parse_mode(parse_mode: str | ReasoningParseMode) -> Rea
     try:
         return ReasoningParseMode(parse_mode)
     except ValueError:
-        logger.warning("未识别的推理解析模式 %s，已回退为 auto", parse_mode)
+        logger.warning(f"未识别的推理解析模式 {parse_mode}，已回退为 auto")
         return ReasoningParseMode.AUTO
 
 
@@ -121,7 +121,7 @@ def _normalize_tool_argument_parse_mode(parse_mode: str | ToolArgumentParseMode)
     try:
         return ToolArgumentParseMode(parse_mode)
     except ValueError:
-        logger.warning("未识别的工具参数解析模式 %s，已回退为 auto", parse_mode)
+        logger.warning(f"未识别的工具参数解析模式 {parse_mode}，已回退为 auto")
         return ToolArgumentParseMode.AUTO
 
 
@@ -425,7 +425,7 @@ def _log_length_truncation(finish_reason: str | None, model_name: str | None) ->
         model_name: 上游返回的模型标识。
     """
     if finish_reason == "length":
-        logger.info("模型%s因为超过最大 max_token 限制，可能仅输出部分内容，可视情况调整", model_name or "")
+        logger.info(f"模型{model_name or ''}因为超过最大 max_token 限制，可能仅输出部分内容，可视情况调整")
 
 
 def _coerce_openai_argument(value: Any) -> Any | Omit: