Allow disabling time auto-advance behavior in tests

[Dentosal]

Closes #4522. See that issue for a more extensive approach.

Motivation

Some programs have background tests that use sleeping for timings while working with IO. Testing such programs with auto-advancing time is currently quite complicated, as it's not possilbe to time the IO correctly..

Solution

This PR adds a way to disable the auto-advance behavior, so that tests can directly manage time using time::advance. It's simply a boolean flag, which is checked when time would be auto-advanced otherwise.

[Darksonn]

There's already a very similar PR open in #5115.

[jorendorff]

As this comment suggests, it's possible to think of the two PRs as independent. This PR seems valuable regardless of whether #5115 is accepted.

It won't help my use case though, because my goal is to use auto-advance, not turn it off!

For our code, I think manually advancing time will prove too tricky. In general, a blocking task, sleep, or timeout could be an implementation detail of any async method, right? So the top-level test code would have to see through all abstractions in order to "know" when to advance time and by how much.

I can't be sure #5115 by itself is up to the task either, but I think letting individual tasks momentarily disable auto-advance is the way to make auto-advance useful in nontrivial cases.

[Darksonn]

Sure, the PRs can be thought of as independently, but the implementations are quite similar.

[madadam]

It seems #5115 already implements all the necessary functionality, it just doesn't expose it to the users of tokio. So all that this PR needs to do is to expose it.

In the meantime, a workaround can be to spawn a dummy blocking task for the time during which one desires to inhibit auto-advance:

let (tx, rx) = mpsc::oneshot();
task::spawn_blocking(move || { rx.blocking_recv() });
// auto-advance is inhibited here
tx.send(());
// auto-advance is enabled again

[Darksonn]

Ah, yeah, I paused looking at this PR due to that other one, but its merged now.

[Darksonn]

This PR has gotten quite out of date. I'm sorry about that. I'll close it for now, but if you want to continue working on this, then feel free to reopen.