STOP_FLUSH race condition

Previous versions of JBoss EAP 5 had a race condition that manifested in some instances of a new node joining a cluster. The race was encountered when a Master node initiated a flush of existing cluster members upon receipt of a join request and then allowed the new node to join the cluster. The new node would then initiate a flush before transferring the initial cluster state. But in some circumstances it could process the stop flush message from the previous flush by the master and return prematurely. Fix: When starting a flush while joining the cluster, any previously received stop_flush notifications are now discarded.