As we discovered while implementing SERVER-119609 we cannot perform a simple recovery from disk as the recovery timestamp may coincide with an invalid state representation on disk for a collection that has acquired the critical section.

As a result, the recovery must be delayed until after the critical section is released which may race with a subsequent oplog 'c' entry on the CSS while recovery is taking place.

This ticket is about implementing the recovery functionality such that oplog replication can enqueue c entries to be applied while disk recovery is taking place and the recovery can apply them in a batch fashion at the end in order to commit the current state to the CSS.