fix(responses): validate litellm params without unpacking untyped kwargs

Adding xai_oauth_token_file to GenericLiteLLMParams adds one unknown
argument at every GenericLiteLLMParams(**kwargs) call site. Building the
params in responses() with model_validate, which behaves the same for a
model without a custom __init__, keeps the basedpyright total below base.

Co-authored-by: Cursor <cursoragent@cursor.com>

litellm/responses/main.py

@@ -1223,7 +1223,7 @@ def responses(
         )
 
         # get llm provider logic
-        litellm_params: Final = GenericLiteLLMParams(**kwargs)
+        litellm_params: Final = GenericLiteLLMParams.model_validate(kwargs)
         try:
             effective_extra_headers: Final = (
                 add_provider_affinity_header(