Bucket not found/unsupported can deadlock starting the http poller

Note: this affects the HEAD of master only and no previous release, introduced during faster failover changes. When the user specifies a non-existent bucket or a memcached bucket there is an edge case race which can lead to a deadlock. The bootstrap process tells the poller controller to stop the cccp poller. The cccp poller has managed to send and receive its own request and has come to the same conclusion. Before Stop is called on cccp poller the poller has already existed. This can lead to Stop being called on a stopped poller, which then waits on a channel that will not be closed.