fix: handle empty endpoints during cloud function reuse (#17501)

Fixes an issue where reusing recently created Cloud Functions during
Remote Function creation would sometimes result in an empty endpoint
(`endpoint=""`).

When a Cloud Function has been recently provisioned, its endpoint URI
(`response.service_config.uri`) may occasionally be returned as an empty
string if URL propagation is still pending. The remote function
decorator previously only checked for `None`, proceeding to create the
BigQuery Remote Function with `OPTIONS(endpoint='')`, which leads to
validation failures when queries invoke the function.

This PR changes the check to `if not cf_endpoint:` to handle both `None`
and empty strings. If the endpoint is empty, it will route to
`create_cloud_function`, which catches the `AlreadyExists` exception and
safely retries/waits for the endpoint URI propagation to complete.

Fixes #<525124882> 🦕

Author: shuoweil

packages/bigframes/bigframes/functions/_function_session.py

@@ -592,7 +592,9 @@ def wrapper(func):
             if reuse is not None:
                 cf_endpoint = self._function_client.get_cloud_function_endpoint(cf_name)
 
-            if cf_endpoint is None:
+            # If the endpoint is empty, the function might exist but the URL propagation is pending.
+            # Running create_cloud_function will handle AlreadyExists and retry endpoint fetching.
+            if not cf_endpoint:
                 cf_endpoint = self._function_client.create_cloud_function(
                     cf_name, cloud_func_spec
                 )