Hot Rod: share the server list globally during failover

Right now we keep list separate topologies for each cache in TopologyInfo. Suppose we have an active cache (many requests going there) and inactive (request once in a couple of hours). During complete cluster update (when all old servers are stopped and new ones booted) the active cache will get the server list updated, but the inactive cache is not informed about this.

With current protocol we cannot piggyback information about server liveness and the cache might be just stopped - therefore we cannot simply remove the server from the other caches.