refactor: improve _wait_for_repo_ready per CodeRabbit review

Improvements:
- Add comprehensive docstring explaining the race condition fix
- Use time.monotonic() for stricter wall-clock timing
- Narrow exception handling to HfHubHTTPError instead of bare Exception

Co-authored-by: CodeRabbit <support@coderabbit.ai>

Author: The-Obstacle-Is-The-Way

tests/test_upstream_hub.py

@@ -268,15 +268,30 @@ def test_push_dataset_dict_to_hub_with_multiple_commits(self, temporary_repo):
             assert num_commits_after_push - num_commits_before_push > 1
 
     def _wait_for_repo_ready(self, repo_id, max_wait=30):
-        """Wait for repository to be in a consistent state."""
-        for _ in range(max_wait):
+        """Wait for repository to be in a consistent state after push operations.
+
+        This helper addresses race conditions where rapid successive push_to_hub calls
+        don't wait for Hub's LFS object propagation between pushes, causing errors like:
+        "LFS pointer pointed to a file that does not exist"
+
+        Args:
+            repo_id: The repository ID to check.
+            max_wait: Maximum time in seconds to wait for repository readiness.
+
+        Raises:
+            TimeoutError: If repository is not ready within max_wait seconds.
+        """
+        from huggingface_hub.errors import HfHubHTTPError
+
+        start_time = time.monotonic()
+        while (time.monotonic() - start_time) < max_wait:
             try:
                 # Verify we can list files (repo is consistent)
                 self._api.list_repo_files(repo_id, repo_type="dataset", token=self._token)
-                # Small delay to ensure LFS objects are propagated
+                # Small delay to ensure LFS objects are fully propagated
                 time.sleep(1)
                 return
-            except Exception:
+            except HfHubHTTPError:
                 time.sleep(1)
         raise TimeoutError(f"Repository {repo_id} not ready after {max_wait}s")