Which approach enables a data engineer to deduplicate incoming records against previously processed data when inserting into a Delta table, in addition to handling intra-batch deduplication for late-arriving records?