Keep track of cluster groups for evacuations

[stgraber]

No description provided.

[stgraber]

The fact that we weren't running cluster requests in notify mode is a bit odd considering the number of places in this code path which was checking whether we were handling a notification (to avoid running config checks twice).

In my local tests, the logic seems to be behaving just fine when in notify mode, so hopefully it was in fact a long standing mistake and not something that we had to do on purpose and failed to document.