Currently the server selector unconditionally sleeps for a certain time and then checks whether there is a suitable server. Instead of this we can have a semaphore-based implementation where the monitoring threads will broadcast any time a server description changes (technically, once the server description has gone through SDAM flow and updated the topology) and the server selector will wait on that semaphore and be immediately awoken when there is a change in topology.

This is split from 1547 because the two notification directions proved to be implementable independently.