Defer requests import in PageIterator to reduce cold import time

Importing requests at module level pulls the entire requests package
(plus charset_normalizer) into every `from msgraph import
GraphServiceClient`, adding ~350 ms (~20-25%) to cold import in an SDK
whose transport is httpx. requests is also not a declared dependency of
this package. Import InvalidURL lazily at the single raise site instead;
exception type and behavior are unchanged.

Author: Lamg222

src/msgraph_core/tasks/page_iterator.py

@@ -11,7 +11,7 @@
  server, and the PageResult class to represent the pages.
 
 This module also imports the necessary types and exceptions from the
-typing, requests.exceptions, kiota_http.httpx_request_adapter,
+typing, kiota_http.httpx_request_adapter,
  kiota_abstractions.method, kiota_abstractions.headers_collection,
 kiota_abstractions.request_information, kiota_abstractions.serialization.parsable,
 and models modules.
@@ -25,7 +25,6 @@
 from kiota_abstractions.request_adapter import RequestAdapter
 from kiota_abstractions.request_information import RequestInformation
 from kiota_abstractions.serialization import Parsable, ParsableFactory
-from requests.exceptions import InvalidURL
 
 from msgraph_core.models.page_result import (
     PageResult,  # pylint: disable=no-name-in-module, import-error
@@ -201,6 +200,9 @@ async def fetch_next_page(self) -> Optional[Union[T, PageResult]]:
         if not next_link:
             raise ValueError('The response does not contain a nextLink.')
         if not next_link.startswith('http'):
+            # Imported lazily: importing requests at module level adds ~350 ms
+            # to every `from msgraph import GraphServiceClient` cold start.
+            from requests.exceptions import InvalidURL
             raise InvalidURL('Could not parse nextLink URL.')
         request_info = RequestInformation()
         request_info.http_method = Method.GET