When a node goes on/offline ensure that the cluster map is up to date

When a node crashes or otherwise goes offline, the client will continue to use the stale configuration. If the node becomes active again, the client will be in an inconsistent state because the node will be flagged as dead and unusable. This commit ensures that the client gets a fresh configuration in the case of a NMV or certain IO errors (forcibly closed or reset) even if the revision hasn't changed.