There's a situation that can occur when using the TerminateAfterJob mode of the agent, where when the ASG gets down to its MinSize, after running a job, the agent will attempt to self-terminate using the `aws autoscaling terminate-instance-in-autoscaling-group` command, but becuase that action would cause the ASG to go below its minsize, the call fails. This fails the entire systemd `ExecPostStop` hook, which fails the entire unit, which causes systemd to restart the unit.

This means that in certain circumstances (when the instance attempted to self-terminate but the call failed), single-job instances can be caused to retain state from previous jobs, which is the opposite of what we want for these instances.

This PR changes the autoterminate script so that if the instance is in TerminateAfterJob mode, if the call to terminate fails, the instance will call `shutdown` on itself, causing the instance to be removed from the ASG no matter what EC2 has to say about it

… true

Then it will appear in the job logs, so customers will know it is set by
the elastic stack.

The latter affects every subsequently started service (at least). The
former only affects the service that needs the env variable.

… Windows

…estart it if needed