Reduce gc_grace_period for all conversation related tables to 1 day. This will
help restart the postgresql migration after a day, if it fails mid way. Lowering
it too much runs the risk of offline nodes resurrecting deleted data.