How would you handle missing or invalid data in a pipeline?