Rebalancing 2.0

Rebalancing needs an overhaul, to incorporate the lessons we have learned.

This is the parent task.

Issues to consider:

• Replicas should be part of replication, at all times replicas need to be available of all data. Access to data should be minimized (i.e. same item should not be accessed more than once during rebalance/replication from one node)
• Client behaviour might need to be better defined, and clients should get more help to avoid unnecessary vBucket errors during rebalance
• Rebalancing needs to consider clusters where not all nodes are up but rebalancing still needs to occur