Detect sbt thread cleaner and terminate workers

[djspiewak]

Also added a non-fatal test which doesn't reproduce the issue. I wasn't able to come up with a good way of rigorously testing this.

Fixes #2704

[armanbilge]

It's super annoying, but you have to register the new example app here for JS:

cats-effect/tests/js/src/main/scala/catseffect/examplesplatform.scala

Lines 31 to 44 in 63fe08c

	object JSRunner {
	val apps = mutable.Map.empty[String, IOApp]
	def register(app: IOApp): Unit = apps(app.getClass.getName.init) = app
	register(HelloWorld)
	register(Arguments)
	register(FatalError)
	register(Canceled)
	register(GlobalRacingInit)
	register(ShutdownHookImmediateTimeout)
	register(LiveFiberSnapshot)
	register(FatalErrorUnsafeRun)
	register(Finalizers)

[vasilmkd]

This clashes with the shutdown() method on WorkStealingThreadPool. I wonder if that method should be called instead, there's already a CAS which will make sure to call it once. Only setting done won't clean up all resources.

[vasilmkd]

The problem with all this is that I don't know if the rest of the thread pools in the same IORuntime are shut down. Technically, IORuntime#shutdown() can also be called, since the WSTP has a reference to its IORuntime.

[djspiewak]

I can swap it to call shutdown(). I'm actually not as worried about the other pools in the runtime because this is such an odd scenario. It'a kind of like being shut down from the inside out, and we can assume it's happening to everyone and not just us.

[vasilmkd]

Thanks for taking this on.