Consider making isolate/rejoin reentrant in Otter ChaosBot

Currently, if multiple experiments run in parallel, isolate/rejoin interaction will break the test. Let's assume that test A isolates node 1 for 1 minute, and test B isolates node 1 for 2 minutes. Expectation is that it should isolate the node for 2 minutes. Instead what happens is that

* test A isolates node 1 (it is isolated)
* test B isolates node 1 (it is still isolated)
* test A rejoins node 1 (it is now functioning)
* test B rejoins node 1 (system breaks with report that node 1 is not isolated)

By having some kind of counter for number of isolations, these experiments could run in parallel without breaking each other.

Start by locating Otter ChaosBot's isolate/rejoin implementation and any tests for parallel experiments. Trace the lifecycle in the issue's two-experiment scenario; done means overlapping isolations keep the node isolated until the final rejoin and no rejoin reports that it is not isolated.

Written by the indexing model from the issue text.