dev: Optimize counting of unconfirmed

[dorjesinpo]

First step in optimizing shutdown logic. The goal of which is to avoid iterating subStreams (preparing for Reliable Broadcast).

The ClusterQueueHelper::d_counterOfUnconfirmed is all we will need for shutdown in the next step. All other counters will (gradually) retire.

[dorjesinpo]

Have some comments, the most important one, are we sure that the loop over downstreams is actually slow?

For the ReliableBroadcast, yes. Consider thread switching too. AtomicInt seems to be a small price for replacing it.

[dorjesinpo]

Reverting to the old way of counting. A relevant PR is https://github.com/bloomberg/blazingmq/pull/399