refactor(proxy): type TokenCountRequest messages and contents as main does

Backport of the TokenCountRequest field types from #47657 to lts/1.108.x, which the #50710 picks rely on to pass the type gate
Hand-ported from ea6dbd2109 (main). Type-only, so request validation and the API schema stay the same

litellm/proxy/_types.py

@@ -3400,12 +3400,12 @@ class LiteLLM_SpendLogs_ResponseObject(LiteLLMPydanticObjectBase):
 class TokenCountRequest(LiteLLMPydanticObjectBase):
     model: str
     prompt: str | None = None
-    messages: list[dict] | None = None
+    messages: list[dict[Any, Any]] | None = None
     """
     Anthropic token counting endpoint uses /messages
     """
 
-    contents: list[dict] | None = None
+    contents: list[dict[Any, Any]] | None = None
     """
     Google /countTokens endpoint expects contents to be a list of dicts with the following structure:
     """