Improve deep-scrub performance with many snapshots
This is accomplished by improving the scrub boundary requirements, so that we don't have a scrub an entire hash value. Now we never need to exceed the osd_scrub_chunk_max value which was causing performance degradation.
Due to the transition to bitwise object sorting we have to handle converting the ScrubMap which was suing BitWise sorting even if NibbleWise is currently in use. Note that the hammer backport will not need this part of the change.