Also some refactoring to make this easier in the future, steps towards
unifying production and testing definitions, and start of docs for the whole thing.

Also, a hostId as required by zfs.

And account for the odd-host-out. :/  See if you can spot it.

The layer above will have to make sure that it is.

There will be no staging grid and the other two nodes in this grid went
elsewhere.

Define testing grid and production grid with a function they can share.

There is not going to be a staging grid.
There will be a testing grid and a production grid.
The testing grid will be loosey goosey.
The production grid will have incremental rollouts.