Restart sleeping Spaces during provisioning

When provisioning a deployment to an existing Space that is STOPPED
or PAUSED, the deployer now automatically restarts it. This ensures
that deployments work correctly even when reusing Spaces that have
been put to sleep by HuggingFace's auto-sleep mechanism.

Key changes:
- Check Space runtime state when updating existing Space
- Call restart_space() API if Space is STOPPED or PAUSED
- Add logging to indicate when Space is being restarted

This fixes the bug where deployments would fail silently when the
target Space was in a sleeping state.

Author: claude

src/zenml/integrations/huggingface/deployers/huggingface_deployer.py

@@ -303,13 +303,22 @@ def do_provision_deployment(
         )
 
         try:
+            from huggingface_hub import SpaceStage
             from huggingface_hub.errors import HfHubHTTPError
 
             # Create Space if it doesn't exist, or update visibility if needed
             try:
                 space_info = api.space_info(space_id)
                 logger.info(f"Updating existing Space: {space_id}")
 
+                # Check if Space is sleeping/paused and restart it
+                runtime = api.get_space_runtime(repo_id=space_id)
+                if runtime.stage in [SpaceStage.STOPPED, SpaceStage.PAUSED]:
+                    logger.info(
+                        f"Space {space_id} is {runtime.stage}. Restarting..."
+                    )
+                    api.restart_space(repo_id=space_id)
+
                 # Update visibility if changed
                 if space_info.private != settings.private:
                     logger.info(