replayer: Add missing flush before committing Kafka offsets
Usually we would add this flush() call at the end of process_replay_objects, but this would make metrics less representative of reality because most of the run time would be in the flush() and not segregated by object type.
Calling flush() at the end would effectively do the same thing as calling it after every object type, so this should not affect performance
This is the proper fix for the issue that swh/infra/ci-cd/swh-charts@59ce8820 works around