Cluster-wide RPCs not ForkChannel friendly

Since this commit:
https://github.com/infinispan/infinispan/commit/f0fb26d43f0176c165bd10134be7b5feed57c893
... Infinispan is no longer usable in WildFly 9, whose transports are configured to use ForkChannel rather than JChannels. The underlying issue is described in JGRP-1905. The commit above adds a cluster-wide RPC on view change (from org.infinispan.topology.ClusterTopologyManagerImpl.confirmMembersAvailable(), the message for which inevitably arrives on the newly starting node before the requisite fork channel is registered. The RPC eventually times out since Infinispan's RPC dispatcher will wait for a response that will never come.