refactor: 用 EndpointFetchConfig 纯数据类替代 ORM 对象传递，统一上游模型缓存管理

- 引入 EndpointFetchConfig dataclass 替代直接传递 ProviderEndpoint ORM 对象，
  避免 DB session 关闭后 DetachedInstanceError
- 新增 build_format_to_config() 统一构建 api_format -> EndpointFetchConfig 映射
- KeyAllowedModelsDialog 改用 useUpstreamModelsCache composable 管理上游模型获取
- useUpstreamModelsCache 增加 error 字段透传部分格式获取失败的 warning
- 删除废弃的 queryProviderUpstreamModels API 函数
- ProviderCandidate 添加 __lt__ 方法支持排序比较

src/api/admin/provider_query.py

@@ -27,7 +27,9 @@ from src.services.model.fetch_scheduler import (
     set_upstream_models_to_cache,
 )
 from src.services.model.upstream_fetcher import (
+    EndpointFetchConfig,
     UpstreamModelsFetchContext,
+    build_format_to_config,
     fetch_models_for_key,
     get_adapter_for_format,
 )
@@ -171,11 +173,8 @@ async def query_available_models(
     if not provider:
         raise HTTPException(status_code=404, detail="Provider not found")
 
-    # 构建 api_format -> endpoint 映射
-    format_to_endpoint: dict[str, ProviderEndpoint] = {}
-    for endpoint in provider.endpoints:
-        if endpoint.is_active:
-            format_to_endpoint[endpoint.api_format] = endpoint
+    # 构建 api_format -> EndpointFetchConfig 映射（纯数据，不依赖 ORM session）
+    format_to_endpoint = build_format_to_config(provider.endpoints)
 
     if not format_to_endpoint:
         raise HTTPException(status_code=400, detail="No active endpoints found for this provider")
@@ -327,7 +326,7 @@ def _aggregate_models_by_id(models: list[dict]) -> list[dict]:
 async def _fetch_models_for_single_key(
     provider: Provider,
     api_key_id: str,
-    format_to_endpoint: dict[str, ProviderEndpoint],
+    format_to_endpoint: dict[str, EndpointFetchConfig],
     force_refresh: bool,
 ) -> Any:
     """获取单个 Key 的模型列表"""