Stricter pytest

[euri10]

It seems to me we could benefit from a stricter pytest, especially the pytest.PytestUnhandledThreadExceptionWarning since we use threads a lot.

Moreover, test exceptions raised in a separate thread dont seem to be handled well with pytest, we have an example of this in the ws test suite, see #760 for instance.
This is also one of the reason our test suite behave strangely as sometimes the whole suite passes fine but seperate tests triggered on their own can fail.

With this pytest configuration we can see 2 issues, fixed after:

1. the test_config left an unclosed socket when the bind_socket() method was called
2. and the whole ws test suite failed with those stricter testing, as the context manager to run the ws server was not closed

this is the error we would get with the stricter config, without the 2 fixes.

>               warnings.warn(pytest.PytestUnraisableExceptionWarning(msg))
E               pytest.PytestUnraisableExceptionWarning: Exception ignored in: <socket.socket fd=-1, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0>
E               
E               Traceback (most recent call last):
E                 File "/home/lotso/PycharmProjects/uvicorn/venv/lib/python3.8/site-packages/_pytest/python.py", line 183, in pytest_pyfunc_call
E                   result = testfunction(**testargs)
E               ResourceWarning: unclosed <socket.socket fd=17, family=AddressFamily.AF_INET, type=SocketKind.SOCK_STREAM, proto=0, laddr=('127.0.0.1', 8000)>

../venv/lib/python3.8/site-packages/_pytest/unraisableexception.py:78: PytestUnraisableExceptionWarning

[euri10]

interestingly the ci fails on windows....on other tests !
have no idea as to why.... will investigate
if you get any idea I'm all ears, dont have a win machine at hand now

[euri10]

will continue on this when I'll have a windows machine, those errors are hard to debug because it's failing on weird places and to be honest the pytest messages are not helping

[euri10]

haha closing this, the test suite changed and the new approach with context manager leads to way less issues it seems, will resurrect when/if #918 lands