HA verify that cluster can be upgraded one node at a time.

The intent is that from 0.24 on a HA cluster can be upgraded node-by-node.
As part of the release process for 0.26 onwards, we need to verify that this does indeed work, and document any restrictions or special proceedures needed to make it work.

We need to verify upgrade from say v1 to v2 works on an active HA cluster:

• start v1 cluster.
• run clients that are active during the upgrade.
• shut down nodes one by one and restart with v2.
• verify no message loss, clients fail-over OK.

We will assume that new features introduced by v2 are not used until the entire cluster is upgraded to v2.

Ideally we want an automated upgrade test, if not we need to document in detail the procedure for a manual test.