Lock manager is not fair in the presence of compatible requests which can be granted immediately

There is code in the lock manager, which attempts to increase system throughput in the case of long-running operations holding intent locks.

This code has the side effect of starving the 'strong' locks (S and X) when there is high load and an interleave of intents on the granted list, which is what would happen under a normal stream of requests. This has caused a bug where replication step down won't complete within the desired timeout.