Fix Signal::CHLD.reset not clearing previously set handler, and fix related race condition in spec

[asterite]

Fixes #7243

I thought about many ways to fix the race condition in this spec... I think the only safe alternative is to use a channel and send that over to the signal handler. If the handler runs before we send the channel, it will wait. If it runs later, the process will be already finished.

The only problem with this spec is that if the CHLD handler doesn't run for some reason (bug) then the spec will wait forever in chan.send because nobody will consume it. However, given that everything is working well now I don't this as a big problem, plus signal handling code doesn't change that frequently.

[ysbaddaden]

Good catch!

[asterite]

Okay... my first commit was missing Signal::CHLD.reset to clear that handler, because later if another spec did Process.new(...) and that process ended then that same handler would be called, hanging everything (on chan.receive), which effectively happened.

I added Signal::CHLD.reset and ran the specs locally... only to find out that that didn't work. The intention of keeping a default handler when Signal::CHLD.reset was good, but any previously set handler wasn't cleared. So this PR also fixes that. Phew!

[ysbaddaden]

Oops, I completely overlooked that. Thanks for fixing this 🙇