Defragmenter task holds HashTable mutexes for longer than necessary

The pause-resume visitor used by the Defragmenter currently holds each hashTable lock for the duration of visiting every HashBucket owned by that lock. This unnecessarily increases the latency of front-end operations attempting to concurrently access the HashTable.

By releasing the hashTable lock after visiting each hash_bucket we could significantly reduce the effect the Defragmenter task has on front-end latency.