Clean up discovered node registry when last node in a cluster crashes

When the last node in a cluster crashes, there is no server-side process to inform the client that not only has a cluster member crashed, but also that the cluster itself has no members. In previous releases, when the last member left the cluster under non-crash circumstances, a CLUSTER_REMOVED message would be sent by the last node to the client to clean up both the node and the cluster entries in the discovered node registry (DNR). In the current implementation, there is no such message sent. See the discussion in WFLY-11682 which also includes a PR solution to the problem in the case that the last node in the cluster shuts down cleanly.

However, we are still left with the case where the last member crashes and does not get a chance to send such a message. In this case, we have entries in the DNR which do not correspond to actual servers and will cause exceptions to be raised when connections are attempted to nodes which do not exist. Blacklisting helps on a per-invocation basis, but there is no mechanism to remove these nodes entirely.