Change loop fixture

[drew2a]

This PR fixes #7126 by it removing the following asyncio-aiohttp compatibility patch from c8c307d:

[kozlovsky]

To my understanding, there are three independent changes in this PR:

• First, it makes some tests non-async.
• Second, it reverts Using SelectorEventLoop during tests #6527 that was implemented to fix Using Proactor event loop on Windows results in random test fails #6456.
• Third, it removes the following asyncio-aiohttp compatibility patch from c8c307d:

@pytest.fixture
def loop(event_loop):
    """
    _This_ fixture masks the original "loop" fixture from pytest-asyncio,
    effectively replacing it with the "event_loop" fixture from aiohttp.
    It solves the following problem:
    pytest-asyncio provides "loop" fixture for creating the event loop,
    aiohttp provides "event_loop" fixture for creating the event loop,
    if you use the "@pytest.mark.asyncio" decorator on a test, it will automatically run the "loop"
    fixture, which could result in test failure if the test uses Futures created with the fixtures
    that use the "event_loop" fixture.
    """
    return event_loop

As there is no explanation in PR, it is hard to say why it is including these changes, and do we really need all three changes at once? Maybe we can split changes into three independent PRs and see how each of them affects tests stability?

I'm a bit concerned about removing the asyncio-aiohttp compatibility patch; it may potentially break aiohttp-related tests in a surprising way.

So, can we split this PR into a three smaller ones and check each of them independently?

[drew2a]

@kozlovsky

we can split changes into three independent PRs and see how each of them affects tests stability.

Yes, we can. Normally I'm trying to do as atomic PR as possible. But it is not always necessary.

This PR indeed combined three changes that should improve test stability.

The reason to combine them together is to decrease the total time on stabilizing our test suit. There are a lot of stability issues and only one developer works on fixing them.

[drew2a]

I'm going to split this PR into three today.

[kozlovsky]

This change can break aiohttp tests in a subtle way, so we need to be careful and be ready to revert it back if some new instabilities arise after the merge.