CI: ban coverage 6.3 that may be causing random hangs in fork test

[tacaswell]

PR Summary

About 24 hours ago (some time 2022-01-25) we started getting random failures on CI that look like:

__________________________________ test_fork ___________________________________
[gw0] linux -- Python 3.7.12 /opt/hostedtoolcache/Python/3.7.12/x64/bin/python

    @pytest.mark.skipif(not hasattr(os, "register_at_fork"),
                        reason="Cannot register at_fork handlers")
    def test_fork():
        _model_handler(0)  # Make sure the font cache is filled.
        ctx = multiprocessing.get_context("fork")
        with ctx.Pool(processes=2) as pool:
>           pool.map(_model_handler, range(2))

lib/matplotlib/tests/test_font_manager.py:206: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
/opt/hostedtoolcache/Python/3.7.12/x64/lib/python3.7/multiprocessing/pool.py:623: in __exit__
    self.terminate()
/opt/hostedtoolcache/Python/3.7.12/x64/lib/python3.7/multiprocessing/pool.py:548: in terminate
    self._terminate()
/opt/hostedtoolcache/Python/3.7.12/x64/lib/python3.7/multiprocessing/util.py:224: in __call__
    res = self._callback(*self._args, **self._kwargs)
/opt/hostedtoolcache/Python/3.7.12/x64/lib/python3.7/multiprocessing/pool.py:617: in _terminate_pool
    p.join()
/opt/hostedtoolcache/Python/3.7.12/x64/lib/python3.7/multiprocessing/process.py:140: in join
    res = self._popen.wait(timeout)
/opt/hostedtoolcache/Python/3.7.12/x64/lib/python3.7/multiprocessing/popen_fork.py:48: in wait
    return self.poll(os.WNOHANG if timeout == 0.0 else 0)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = <multiprocessing.popen_fork.Popen object at 0x7f258326c450>, flag = 0

    def poll(self, flag=os.WNOHANG):
        if self.returncode is None:
            try:
>               pid, sts = os.waitpid(self.pid, flag)
E               Failed: Timeout >300.0s

/opt/hostedtoolcache/Python/3.7.12/x64/lib/python3.7/multiprocessing/popen_fork.py:28: Failed
----------------------------- Captured stderr call -----------------------------

~~~~~~~~~~~~~~~~~~~~~ Stack of <unknown> (139800230110976) ~~~~~~~~~~~~~~~~~~~~~
  File "/opt/hostedtoolcache/Python/3.7.12/x64/lib/python3.7/site-packages/execnet/gateway_base.py", line 285, in _perform_spawn
    reply.run()
  File "/opt/hostedtoolcache/Python/3.7.12/x64/lib/python3.7/site-packages/execnet/gateway_base.py", line 220, in run
    self._result = func(*args, **kwargs)
  File "/opt/hostedtoolcache/Python/3.7.12/x64/lib/python3.7/site-packages/execnet/gateway_base.py", line 967, in _thread_receiver
    msg = Message.from_io(io)
  File "/opt/hostedtoolcache/Python/3.7.12/x64/lib/python3.7/site-packages/execnet/gateway_base.py", line 432, in from_io
    header = io.read(9)  # type 1, channel 4, payload 4
  File "/opt/hostedtoolcache/Python/3.7.12/x64/lib/python3.7/site-packages/execnet/gateway_base.py", line 400, in read
    data = self._read(numbytes - len(buf))
------------------------------ Captured log call -------------------------------
DEBUG    matplotlib.axes._base:_base.py:2965 title position was updated manually, not adjusting
DEBUG    matplotlib.backends.backend_pdf:backend_pdf.py:875 Assigning font /b'F1' = '/home/runner/work/matplotlib/matplotlib/lib/matplotlib/mpl-data/fonts/ttf/DejaVuSans.ttf'
DEBUG    matplotlib.backends.backend_pdf:backend_pdf.py:917 Embedding font /home/runner/work/matplotlib/matplotlib/lib/matplotlib/mpl-data/fonts/ttf/DejaVuSans.ttf.
DEBUG    matplotlib.backends.backend_pdf:backend_pdf.py:924 Writing TrueType font.

We are seeing this mostly on py37, but it also showed up on 3.8 / 3.9 so this won't just go away when we merge #22194 .

The only version change between the last working run to the first failing run was coverage 6.2 -> 6.3. To get coverage on subprocesses there is some extra configuration done at (sub-)process spawn time to get the hooks in place (which is actually managed for us by pytest-cov) so it is plausible that this is related, but do not actually understand the underlying problem.

[QuLogic]

Not entirely sure why this broke things, but I see the change in dependencies in the Actions logs, as you say.

[QuLogic]

Might be an upstream issue here: nedbat/coveragepy#1310