Connect to different zookeeper clusters async

We've run into a few situations where hitless nerve reloads aren't so hitless because we have 5 registrations that take 5 seconds (zk timeout), and then our 30s sleep isn't enough.

Perhaps we should make it async.

Start by tracing how the five ZooKeeper registrations are performed during a nerve reload and where the 30-second sleep is applied. Reproduce the reported five-registration, five-second-timeout scenario, then define completion as reloads no longer waiting serially or exceeding the available timeout; the issue does not name files or tests.

Written by the indexing model from the issue text.