Get rid of the ParallelBatchWriter mutex (aka "peanut-butter" lock)

The ParallelBatchWriter lock needs to be acquired by each thread in both standalone and replicated instances. Now that the MMAP V1 journaling is split into global and flush locks, we can get rid of the PBR lock.

The difference would be that the replication apply log coordinator will take the global lock in X mode before spawning the applier threads. Each applier thread then will not take the global lock (but will still take the flush lock in MMAP V1).