Fix middleware traceback fetching on Python 3.8+, fix ResourceWarnings in TestClient, fix CI build

[JayH5]

This would ideally be 3 separate PRs, but all of these issues have broken the build on master so it'd be difficult to demonstrate them working separately. Still, let me know if you'd like me to split things up.

The issues are:

1. Mypy 0.800 has some changes w.r.t module discovery. The mypy run was failing because mypy wanted the test functions in tests/middleware/* to have type annotations, despite this being disabled for tests submodules. Mypy was not correctly identifying that these tests were submodules due to the lack of an __init__.py file. See somebody having a similar problem in mypy.ini module-specific override doesn't work without __init__.py python/mypy#9974. An alternative (or something we could also do) would be to add namespace_packages = True to the config. Since we have __init__.py files everywhere else, I went with that option for now.
2. Something in a dependency (pytest? Python itself?) changed so that we were having a bunch more ResourceWarnings detected due to unclosed sockets. These (Unix) sockets I believe were setup by the asyncio event loop as part of its internal machinery. In the TestClient, in the WebSocketTestSession we weren't closing the event loop used by the worker thread there. I changed the loop creation to be done inside the thread (since it is only used by the thread, and can never be used again once the thread has completed) and closed the loop at the end of its use (also in the thread). Possibly fixes fix Unraisable ResourceWarnings #1050
3. In ServerErrorMiddleware we were using the undocumented exc_traceback attribute of TracebackError. This attribute was removed for bpo42482 in Python 3.9.1 & 3.8.7. I switched to using the __traceback__ attribute of exceptions which I think should have the same value (it at least produces the same HTML for the test case we have). I also added a check that the exception actually had a traceback, which we weren't doing previously but probably should've been. Fixes Test failure #1131, The debug mode does not work well #1126

[euri10]

Something in a dependency (pytest? Python itself?)

It's pytest itself (since 6.2 https://docs.pytest.org/en/stable/usage.html#unraisable) and if you look it happens only on 3.8 and above as expected.
Since we turn warnings into exceptions then this popped off.
Uvicorn has a pending PR to do the same.

[florimondmanca]

Thanks so much ❤️

[florimondmanca]

@JayH5 I updated the title bc there are 2 items we'll want this PR to be included in the changelog for (fixing tracebacks, fixing ResourceWarning instances).

[florimondmanca]

Depending on what's currently in master I guess we can follow up with a bug fix release?

[JayH5]

Thanks @euri10 @florimondmanca ❤️

I'll look at doing a release soon.