Ceph

Currently, there is a race on the replica between processing the repops removing objects from a snap collection and processing the updated info with the snap collection removed. If the latter occurs first, the replica will crash on a not empty error. Waiting on the repop in the snap_trimmer loop, however, would tie up a disk_tp thread. Instead, I am refactoring snap_trimmer into a set of non-blocking operations which will be called from substates of Active.