Given a step with function `foo`, and cleanup function `foo_cleanup`, should `foo_cleanup` be called, if `foo` fails? I think it should, to handle cases where `foo` does some things, but then fails. Ideally `foo` would clean up after itself upon failure, but it'd be more ergonomic to not require that. `foo_cleanup` needs to be written in a way that can it works even if only some of the things it's cleaning up are there. Specific case that triggered this: `foo` starts a virtual machine, but can fail after it's already running. `foo_cleanup` doesn't get run, and as a result the VM is still there after the Subplot-generated test program finishes.