Implement new test dispatchers

[dkhalanskyjb]

Defines two test dispatchers:

• StandardTestDispatcher, which, combined with runTest, gives an illusion of an event loop;
• UnconfinedTestDispatcher, which is like Dispatchers.Unconfined, but skips delays.

By default, StandardTestDispatcher is used due to the somewhat chaotic execution order of Dispatchers.Unconfined. TestCoroutineDispatcher is deprecated.

Fixes #1626
Fixes #1742
Fixes #2082
Fixes #2102
Fixes #2405
Fixes #2462

[qwwdfsad]

I do not recall our final decision regarding this, maybe we've discussed it and decided not to do:

Are we going to add runTestUnconfined for people to gracefully migrate from runBlockingTest and to have a counterpart for runTest?

[qwwdfsad]

Also, we can then bulk-replace deprecated runBlockingTest in our tests

[dkhalanskyjb]

Are we going to add runTestUnconfined for people to gracefully migrate from runBlockingTest

We didn't decide on this in our discussion. However, checking more closely how Unconfined behaves, I discovered that it does not, in fact, enter launch and async blocks eagerly. The only reliable behavior I got from it is that launching a coroutine with Unconfined leads to all dispatches to that coroutine from non-unconfined dispatchers to happen eagerly. Launching the whole test under UnconfinedTestDispatcher leads to haphazard execution. I still decided to keep it given how useful the unconfined dispatcher is when child coroutines are launched with it.

So, it may turn out that we'll need what is now the TestCoroutineDispatcher after all, just with a different name, and probably with slightly different behavior.

Also, we can then bulk-replace deprecated runBlockingTest in our tests

I don't think we should do that. The only tests that use runBlockingTest are the legacy ones that ensure that we didn't break a lot of existing code during the transition. I checked the tests that use runBlockingTest and, for each one that I found useful, I added the corresponding version in terms of the new primitives.

[dkhalanskyjb]

(This is a test to see if I can still write in this discussion; a Github bug could prevent me)

[qwwdfsad]

Maybe it's worth to avoid such caution, especially if we would like to promote standard dispatcher (and, well, runTest in general) as the default one

[qwwdfsad]

Also, it may be good to state that such dispatcher is intended to be used in advanced scenarios and show one with a code example to prevent people from using it in a new code where it is not necessary and to warn that this may be not the best migration path from runBlockingTest

[dkhalanskyjb]

Done.