fix(nox-api): address duplicate creation risks, pending upgrading issues, cache warm error handling, test speedups, and frontend display support for pending keys

backend/app/services/upstream_client.py

@@ -380,12 +380,14 @@ _RATE_LIMIT_BACKOFFS = (2, 5, 10, 20)
 _RATE_LIMIT_MAX_WAIT = 60
 
 
-def _retry_on_429(fn: Callable[[], Any], backoffs: tuple[int, ...] = _RATE_LIMIT_BACKOFFS) -> Any:
+def _retry_on_429(fn: Callable[[], Any], backoffs: tuple[float, ...] | None = None) -> Any:
     """执行 fn()，若遇 429 则按退避序列重试，超限后抛 UpstreamError。
 
     fn 应是 HTTP 调用；429 响应会被 httpx.HTTPStatusError 捕获（raise_for_status 之后）
     或者 fn 内部已经解析为 UpstreamError 且 message 含 429。
     """
+    if backoffs is None:
+        backoffs = _RATE_LIMIT_BACKOFFS
     total_waited = 0
     for attempt, wait in enumerate(backoffs):
         try:
@@ -783,7 +785,7 @@ class UpstreamClient:
 
         缓存以 token name 为 key；同名取最新（id 最大）。
         """
-        all_tokens = self._list_all_new_api_tokens()
+        all_tokens = _retry_on_429(lambda: self._list_all_new_api_tokens())
         cache: dict[str, dict[str, Any]] = {}
         for t in all_tokens:
             name = str(t.get("name") or t.get("key_name") or "")