refactor: 增强跨格式转换系统，支持 thinking/文件/音频/响应格式等完整转换

- 新增 ThinkingConfig/ResponseFormatConfig/FileBlock/AudioBlock 内部表示
- 实现 OpenAI reasoning_effort <-> Claude thinking <-> Gemini thinkingConfig 互转
- 支持 OpenAI web_search_options -> Claude web_search tool 转换
- 新增异步 convert_request_async，在转换阶段解析图片 URL 为 base64
- 将 GlobalModel.output_limit 传播至跨格式转换用于 max_tokens 默认值
- 前端模型表单增加最大输出 Token 和上下文窗口配置
- Claude normalizer 保留 cache_control 透传（system 数组和 content block）
- Gemini normalizer 支持内置工具（googleSearch/codeExecution/urlContext）
- OpenAI normalizer 补全采样参数、response_format、usage details 转换
- 各 normalizer 工具方法提取至基类消除重复代码
- 简化 Kiro model mapping 为直接透传
- 预置模型列表添加 claude-sonnet-4.6

src/api/handlers/base/chat_handler_base.py

@@ -706,11 +706,12 @@ class ChatHandlerBase(BaseMessageHandler, ABC):
         # 跨格式：先做请求体转换（失败触发 failover）
         registry = get_format_converter_registry()
         if needs_conversion:
-            request_body = registry.convert_request(
+            request_body = await registry.convert_request_async(
                 request_body,
                 str(client_api_format),
                 str(provider_api_format),
                 target_variant=cross_format_variant,
+                output_limit=candidate.output_limit if candidate else None,
             )
             # 格式转换后，为需要 model 字段的格式设置模型名
             self._set_model_after_conversion(

src/api/handlers/base/cli_protocol.py

@@ -155,7 +155,7 @@ class CliHandlerProtocol(Protocol):
         mapped_model: str | None,
     ) -> str | None: ...
 
-    def _convert_request_for_cross_format(
+    async def _convert_request_for_cross_format(
         self,
         request_body: dict[str, Any],
         client_api_format: str,
@@ -165,6 +165,7 @@ class CliHandlerProtocol(Protocol):
         is_stream: bool,
         *,
         target_variant: str | None = ...,
+        output_limit: int | None = ...,
     ) -> tuple[dict[str, Any], str]: ...
 
     def _extract_response_metadata(

src/api/handlers/base/cli_request_mixin.py

@@ -228,7 +228,7 @@ class CliRequestMixin:
             stream_options["include_usage"] = True
             request_body["stream_options"] = stream_options
 
-    def _convert_request_for_cross_format(
+    async def _convert_request_for_cross_format(
         self,
         request_body: dict[str, Any],
         client_api_format: str,
@@ -238,6 +238,7 @@ class CliRequestMixin:
         is_stream: bool,
         *,
         target_variant: str | None = None,
+        output_limit: int | None = None,
     ) -> tuple[dict[str, Any], str]:
         """
         跨格式请求转换的公共逻辑
@@ -252,16 +253,18 @@ class CliRequestMixin:
             fallback_model: 备用模型名（通常是原始请求的 model）
             is_stream: 是否流式请求
             target_variant: 目标变体（如 "codex"），用于同格式但有细微差异的上游
+            output_limit: GlobalModel 配置的模型输出上限
 
         Returns:
             (转换后的请求体, 用于 URL 的模型名)
         """
         registry = get_format_converter_registry()
-        converted_body = registry.convert_request(
+        converted_body = await registry.convert_request_async(
             request_body,
             str(client_api_format),
             str(provider_api_format),
             target_variant=target_variant,
+            output_limit=output_limit,
         )
 
         # 先计算 URL 模型（在清理 body 中的 model 字段之前）

src/api/handlers/base/cli_stream_mixin.py

@@ -317,7 +317,7 @@ class CliStreamMixin:
 
         # 跨格式：先做请求体转换（失败触发 failover）
         if needs_conversion and provider_api_format:
-            request_body, url_model = self._convert_request_for_cross_format(
+            request_body, url_model = await self._convert_request_for_cross_format(
                 request_body,
                 client_api_format,
                 provider_api_format,
@@ -325,6 +325,7 @@ class CliStreamMixin:
                 ctx.model,
                 is_stream=upstream_is_stream,
                 target_variant=conversion_variant,
+                output_limit=candidate.output_limit if candidate else None,
             )
         else:
             # 同格式：按原逻辑做轻量清理（子类可覆盖）

src/api/handlers/base/cli_sync_mixin.py

@@ -150,7 +150,7 @@ class CliSyncMixin:
 
             # 跨格式：先做请求体转换（失败触发 failover）
             if needs_conversion and provider_api_format:
-                request_body, url_model = self._convert_request_for_cross_format(
+                request_body, url_model = await self._convert_request_for_cross_format(
                     request_body,
                     client_api_format,
                     provider_api_format,
@@ -158,6 +158,7 @@ class CliSyncMixin:
                     model,
                     is_stream=upstream_is_stream,
                     target_variant=conversion_variant,
+                    output_limit=candidate.output_limit if candidate else None,
                 )
             else:
                 # 同格式：按原逻辑做轻量清理（子类可覆盖）