refactor: 拆分职责、引入 dataclass 封装并增强缓存健壮性

- ErrorClassifier 副作用操作分离为 ErrorHandlerService（缓存失效、健康记录、RPM 调整）
- chat_handler_base 提取 ProviderRequestResult dataclass 和 _prepare_provider_request 方法
- failover 提取 AttemptErrorOutcome dataclass 和辅助方法
- formula_engine 拆分 _resolve_mapping 为子方法，增加求值异常日志
- usage recording 引入 UsageCostInfo dataclass 封装成本参数
- 前端 types.ts 拆分为 types/ 子模块
- cache backend 工厂函数加锁防止并发重复创建，LocalCache 容量检查修正
- CacheSync 监听增加断线重连机制，publish 增加重试
- guide 页面修正 useSiteInfo() 调用顺序

tests/services/test_format_conversion_health_buckets.py

@@ -1,4 +1,6 @@
+from collections.abc import AsyncGenerator
 from contextlib import asynccontextmanager
+from typing import Any
 from unittest.mock import AsyncMock, MagicMock, patch
 
 import pytest
@@ -8,7 +10,7 @@ from src.services.request.executor import RequestExecutor
 
 
 @asynccontextmanager
-async def _noop_async_cm():
+async def _noop_async_cm() -> AsyncGenerator[None]:
     yield
 
 
@@ -46,7 +48,9 @@ async def test_executor_records_health_by_provider_format() -> None:
     candidate.key = key
     candidate.is_cached = False
 
-    async def request_func(_provider, _endpoint, _key, _candidate):  # noqa: ANN001
+    async def request_func(
+        _provider: Any, _endpoint: Any, _key: Any, _candidate: Any
+    ) -> dict[str, bool]:
         return {"ok": True}
 
     with (
@@ -96,7 +100,7 @@ async def test_error_classifier_records_failure_by_provider_format() -> None:
     key.id = "k1"
 
     with patch(
-        "src.services.orchestration.error_classifier.health_monitor.record_failure"
+        "src.services.orchestration.error_handler.health_monitor.record_failure"
     ) as record_failure:
         await classifier.handle_retriable_error(
             error=RuntimeError("boom"),