Allow ! as the return type of proc/closure literals

[ftxqxd]

Fixes #13490.

[alexcrichton]

This may be a very untested part of the compiler (as it may have never been exercised before), so it would be nice to see a larger array of tests to ensure that this feature does indeed work. Some things such as:

• If a ! closure returns, an error should be generated.
• Code after a ! closure invocation should be dead.
• Calling a ! closure in a ! function should be ok

[ftxqxd]

There’s already some testing for this: compile-fail/closure-that-fails.rs tests that closures that return ! cannot return. But I’ll add some more tests to this PR.

[nikomatsakis]

More tests I would like:

• A closure that is declared as -> ! but in fact does not fail (compile-fail).
• A closure being passed to a fn that expects -> !
  • This raises the question of whether we will require an explicit -> ! also in the callee side, I think we will
    • To be honest, I am happy with more-or-less either behavior, but I want a test showing what it is

[ftxqxd]

I think compile-fail/closure-that-fails.rs covers most of those cases, but it doesn’t directly specify the -> ! type in the closure literal, only in the function declaration. I’ll update that test to test that -> ! actually means something in literals.