[NEW] Server event on cluster topology change

The problem/use-case that the feature addresses

Each search query operation must be fanned out to each shard (selecting from primary/replica accordingly). This requires processing of the entire cluster map, which is an expensive process and a substantial amount of the mainthread processing time for a query operation.

Description of the feature

If a notification of a change in the cluster map could be made available to search, then it could easily use a cached copy of the cluster map -- invalidating that cache as needed. Seems like a server event would be the way to go. In other words, if a server event was published any time there was a change in the cluster map.