A data engineer is designing a pipeline to handle late-arriving and duplicate records. Beyond de-duplicating data within the current micro-batch, which technique effectively prevents duplicate records from being inserted into an existing Delta table by checking against previously processed data?