Hard kill tests

[goodboy]

As per some light commentary in #144, we need some hard kill tests for wen processes fail due to internal machinery failure. An easy way to test this now should be to simply cancel one of the actor internal nurseries and ensure we don't get indefinite hangs; containing nursery/supervisor logic should hard kill (via signal) the subprocess after a suitable timeout.

I've seen this work in practice during manual use but we of course want to prevent regressions.