Figure out Jepsen behaviour for unavailable cluster

[MathieuBordere]

Find out if Jepsen reports an error when a theoretically available cluster is unavailable.

[cole-miller]

Just brainstorming a bit about this. I'm imagining a situation like: we do some damage to the cluster, then heal it, and we expect that afterward our requests will start to complete successfully again (availability). I do not think the Jepsen checkers that we're using right now can test for this kind of thing; they're looking for either anomalies in the history of operations or red flags like assertion failures, core dumps, and uncaught exceptions. The availability check, and the circumstances under which we expect a cluster to be available, seem like they'd be pretty specific to each distributed system, and we'll need to write our own logic to implement them for dqlite.