Don't remove the poller controller watcher from config updates

Within the poller controller we have a watcher for when the SDK receives a cluster config. The watcher is responsible for transitioning from using the http poller to the cccp poller if relevant. However we remove this watcher as soon as we see a config.

There are scenarios where we can end up with the http poller being triggered during normal operation of the SDK, not just at startup time. This can occur when a node is being rebalanced out of the cluster and the timing is just right that a node is bootstrapping against it and that node replies to select bucket with a bucket not found error.

Due to this we should leave the watcher running so that we can always transition back to CCCP again.