A lot of our life cycle was getting really clunky and we are frequently seeing tests hang, especially on Windows. This PR aims to improve the situation:

1. Ensure that we join the boot thread when deactivating the add-on. This one is mostly about tests. Tests quickly create new add-on instances, which boots the Rails app. If we don't join the boot thread, we risk finishing the tests before the Rails app is up and we may end up with orphaned processes
2. Got rid of the force_kill approach. This was basically a brute force way of trying to ensure things were shutting down properly
3. Started making sure that the runtime server calls exit!(0) once the main loop finishes, to terminate child processes with it
4. Removed the process group ID stuff, which I don't believe was actually doing anything useful
5. Started handling early termination of stdin in the runtime server, so that it shuts down gracefully if the pipe was closed on the other end