bpo-36668: FIX reuse semaphore tracker for child processes

[tomMoral]

The current implementation of the semaphore_tracker creates a new process for each children.
This PR intends to fix this behavior.

The easy fix would be to pass the _pid to the children but the current mechanism to check if the semaphore_tracker is alive relies on waitpid which cannot be used in child processes (the semaphore_tracker is only a sibling of these processes). The main issue is to have a reliable check that either:

• The pipe is open. This is what is done here by sending a message. I don't know if there is a more efficient way to check it.
• Check that a given pid is alive. As we cannot rely on waitpid, I don't see an efficient mechanism.

I took the approach of adding a PROBE command in the semaphore tracker. When the pipe is closed, the send command fails and this means the semaphore tracker is down.

https://bugs.python.org/issue36668

[tomMoral]

@pitrou This PR avoid spawning semaphore_tracker in each subprocesses.

[tomMoral]

@pitrou I think this PR is ready

[pitrou]

Looks mostly good to me. Just a single comment.

[pitrou]

+1 from me

[pitrou]

@tomMoral Can you force-push to trigger an AppVeyor build?

[tomMoral]

Done

[bedevere-bot]

@pitrou: Please replace # with GH- in the commit message next time. Thanks!

[pitrou]

Thanks @tomMoral . I don't think this needs to be backported to 3.7, do you concur?

[tomMoral]

@pitrou yes I agree this is not critical so it is not necessary to backport to 3.7.