BFM backfill allocation unfair when we hit running backfill limit

Originally seen in MB-44562. We currently unfairly schedule backfills when we hit the max running backfill limit as more frequently running BackfillManagers have more opportunity to acquire that resource.

This patch http://review.couchbase.org/c/kv_engine/+/147424 attempted to address that issue by queueing the BackfillManagers in a unique queue so that they would be scheduled more fairly when hitting the max running limit. This patch hit the issue described in MB-44919 though as BackfillManagers could be destroyed but not removed from the new pendingQueue. The code added is complicated and may be simplified through larger refactorings so the original change was reverted to be addressed later.