Using startup script for dispatcher instead of SSH doesn't throw errors when it should

If you try to start a dispatcher when you don't have enough quota, run_experiment.py will succeed since the instance will be created briefly. But Google Cloud will shut it down.
It's probably a good idea to continue using the startup script but wait until we can SSH into the dispatcher before reporting a success. This still gives us much of the benefits of using a startup script: having run_experiment.py terminate in a reasonable amount of time.
This will break @Shadoom7 but I think the issue on his machines is unique enough that is less of a concern than this problem.