Add a node in ZK to tell all masters to shutdown.

Currently, shutting down a multi-master HBase is a bit of a mess since the backup masters aren't pinged by the stop-hbase script. To shut them down, you have to do a hbase-daemons.sh stop master. Also, with master failover in place, the backup masters may try to come alive. This can be solved by adding a "shutdown node" in ZK on which all masters should be listening.