Query UI caches list of valid nodes too aggressively

When you first visit the query page it computes the list of valid nodes and this cache isn't invalidated sufficiently often.

For instance, on a cluster with no query service, when visited this page will politely inform the user that there's no query service available. When a node with a query service is added it will continue to show the same message.

Pavel stashes the current value of the call to pools/default in the $rootScope - suggest the best way is to not cache the valid nodes but compute it as needed.