Once we allow PIT reads on secondaries within previous batches, we should be able to simply recover to the common point on rollback instead of recovering to the stable timestamp and replaying the oplog to the common point. This will make rollback even faster and easier to understand. The recovery code can then only be used for startup recovery.