Your ML pipeline uses Vertex AI Feature Store to serve features for online predictions. You need to monitor the freshness of features in the online store. Which approach is most effective?