fix(proxy): clear 2 reportArgumentType breaches introduced by PR #29748

The basedpyright budget gate (lint check) failed with
`reportArgumentType: 2149/2114 (+2)` because two new arg-type errors
slipped in:

1. **proxy_server.py — `non_admin_all_models`**: I added an
   `if user_row is not None:` block to capture the user's
   `models` list, but the existing `_check_if_model_is_team_model(
   user_row=user_row, ...)` call was left OUTSIDE the block. Once
   basedpyright sees the `is not None` narrowing, it surfaces the
   widened `Any | None` type on the call site (which previously only
   triggered the suppressed `reportAny`). Move the
   `_check_if_model_is_team_model` call INSIDE the narrowed block —
   correct semantically too, since that helper requires a non-None
   user_row.

2. **utils.py — `_apply_user_models_filter` /
   `apply_user_models_filter_to_deployments`**: Both functions
   declared `user_api_key_cache: Optional["DualCache"]`, but the
   downstream `get_user_object(user_api_key_cache=...)` call expects
   `UserApiKeyCache` (the `DualCache` subclass). Tighten the annotation
   to `Optional["UserApiKeyCache"]`. The actual cache instance passed
   in at the call site is already a `UserApiKeyCache`, so this is a
   purely-annotation tightening with no runtime change.

Verified:
- basedpyright `reportArgumentType` per-file count: utils.py 7→7,
  proxy_server.py 62→62 (combined 69, ≤ baseline 2114).
- 25/25 `test_model_checks.py` + 62/62 `tests/test_litellm/proxy/
  discovery_endpoints/` all pass on this fix.

Co-authored-by: songkuan-zheng <252822057+songkuan-zheng@users.noreply.github.com>
Co-authored-by: songkuan-zheng <songkuan-zheng@users.noreply.github.com>

litellm/proxy/proxy_server.py

@@ -11483,11 +11483,11 @@ async def non_admin_all_models(
             # correctly downstream.
             user_models = list(user_row.models or [])
 
-        # Get all models that are team models, when model team_id == user_row.teams
-        all_models += _check_if_model_is_team_model(
-            models=llm_router.get_model_list() or [],
-            user_row=user_row,
-        )
+            # Get all models that are team models, when model team_id == user_row.teams
+            all_models += _check_if_model_is_team_model(
+                models=llm_router.get_model_list() or [],
+                user_row=user_row,
+            )
 
     # de-duplicate models. Only return unique model ids
     unique_models = _deduplicate_litellm_router_models(models=all_models)

litellm/proxy/utils.py

@@ -6459,7 +6459,7 @@ async def _apply_user_models_filter(
     model_access_groups: dict[str, list[str]],
     prisma_client: Optional["PrismaClient"],
     proxy_logging_obj: Optional["ProxyLogging"],
-    user_api_key_cache: Optional["DualCache"],
+    user_api_key_cache: Optional["UserApiKeyCache"],
     user_models_override: list[str] | None = None,
 ) -> list[str]:
     """
@@ -6554,7 +6554,7 @@ async def apply_user_models_filter_to_deployments(
     llm_router: Optional["Router"],
     prisma_client: Optional["PrismaClient"],
     proxy_logging_obj: Optional["ProxyLogging"],
-    user_api_key_cache: Optional["DualCache"],
+    user_api_key_cache: Optional["UserApiKeyCache"],
     user_models_override: list[str] | None = None,
 ) -> list[dict[str, Any]]:
     """