Capped delete uses fast size and count to check for limits. When the fast size and count drift over time, a large backlog of excessive documents could accumulate. This could lead to a rebuilt node with the correct fast size and count trying to delete these excessive documents all at once.

The deletes now happen as part of the storage transaction the user write has and are unbounded. We should at least impose an upper bound on the amount of data a capped delete should try to do.