The driver presently uses a server's connection pool in order to do monitoring checks. This has well documented limitation, in particular in the face of a default socket timeout of unlimited. We should introduce a Monitor class, and a dedicated monitoring socket for each server in a Topology.