You've developed an updated version of a Dataflow streaming pipeline that ingests data from Pub/Sub into BigQuery. The current production pipeline uses a 5-minute window for processing. How can you deploy the new version without risking data loss, inconsistencies, or exceeding a 10-minute increase in processing latency?