SERVER-29210 moved WiredTiger's checkpointing thread into MongoDB such that it can later be coordinated with replication to only persist to disk majority committed updates (and the entirety of the oplog). Updates that are not majority committed will only modify in-memory data structures. This will make rollbacks simpler by reverting to the on-disk data and rolling forward the updates that survived the rollback.

This ticket is for exposing this functionality across storage engines by (presumably) allowing them to opt-in to some new storage engine API methods and pulling the checkpoint thread up a level.