Reimplement process exit expect failure

[Naatan]

The original implementation gives responsibilities to output consumers that they should have not had. Essentially we ended up with a structure where termtest used output consumers, and output consumers in turn used termtest. This feels error-prone and difficult to reason about.

This rewrite fixes this by sending a Stop signal to consumers that are still pending, when the output producer reaches EOF (ie. process exited). This should also avoid any race conditions wrt consuming output shortly after the process exited, which was the main issue I was trying to fix.

I should've realized this when I reviewed the original PR, but it had been a while since I worked on this code.

[CLAassistant]

All committers have signed the CLA.

[Naatan]

@mitchell-as sorry I had flagged you too early, there were failing tests on windows.

Turns out on Windows the PTY won't reach EOF unless I have instructed the library that I am waiting for it to close. This is a bug in the library, but for now we'll have to live with it. So the recent changes have us start waiting for the pty to close immediately via goroutine.