this doesn't involve too many code changes, more just shifting code
around. the reason for doing this is that, if we can move as many partso
f deploys as possible into an FSM, it lets us more easily resume deploys
that failed, or from midway through a canceled deploy, or anywhere else
in a deploy.

I want to make more granular FSM transitions for each individual machine
too, which I'll either do in a future commit or a future PR. The more
individual FSM states we can retry, the more granular 'resuming' we can
do. If we can move other parts of deploys into an FSM, it also gives us
the ability to resume *those* parts as well. Plus some extra super
powers like easily cloning apps, more granular view of deploy failures
in a way that we can retry locally, declaratively configuring a fly app, etc.