Allow Manual Failover in cluster bootstrap

[arohter]

When launching fresh clusters, when zk holds no 'nodes' data, and all redis nodes are in master state, client.manual_failover() will fail with an exception due to the build_client() call failing to find the 'nodes' data when initializing the client object.

At the moment we need to manually issue slaveof commands to redis nodes in order to bootstrap.

We should be able to just call a static class method (skipping build_clients) to force a bootstrap.