Rug scaling

This proposes allowing multiple RUG processes to be spawned across many nodes. Each RUG process is responsible for a fraction of the total running appliances. RUG_process->appliance(s) mapping will be managed by a consistent hash ring. An external coordination service (ie, zookeeper) will be leveraged to provide cluster membership capabilities, and python-tooz will be used to manage cluster events. When new members join or depart, the hash ring will be rebalanced and appliances re-distributed across the RUG.

This allows operators to scale out to many RUG instances, eliminating the single-point-of-failure and allowing appliances to be evenly distributed across multiple worker processes.