Terminate process group after cram execution

[Leonidas-from-XIV]

After discussing an alternate solution for #11827 with @rgrinberg we agreed that the best course of action is to just terminate the entire process group of the cram test after execution.

This PR reuses @Alizter's cram test repro case.

It introduces a slight race condition that we might be terminating an unrelated process group that accidentally got the same PID as the cram test. I am unaware of a way to avoid it, the chance is not substantial as OSes usually increment PIDs until they wrap around. In any case the engine is already terminating process groups if the process to be waited on has a timeout, so this code just extends it to the case where there is no timeout.

Fixes #11820

[Leonidas-from-XIV]

Surprising behavior on Windows when we kill_process_group which attempts to Unix.kill the PID which somehow causes corruption in the files that ocamldep writes. That's odd because at that time we have awaited the process so it should have exited already.

[Alizter]

@Leonidas-from-XIV that sounds to me like we have a race condition. Is it similar to the issue in #11010?

[rgrinberg]

Seems like there's a failure in CI. Is it something relevant?

[Leonidas-from-XIV]

I'm trying to figure out. It works reliably on my system (only failures are from Melange), CI sometimes hangs forever which is unrelated to this PR but its hard to say whether the failure is related since I can't reproduce it.

[Alizter]

@Leonidas-from-XIV Could you push a temporary commit disabling all of exec-watch and perhaps watching. Those would be my goto culprits for tests that doesn't terminate.