Migrate CI/CD from AppVeyor to GitHub Actions

[clyde-johnston]

GitHub Actions workflow created to replace AppVeyor.

Note: assert() was removed from the _unittest_slow_cli_pub_sub_anon test to make it work on Ubuntu runners.

[pavel-kirienko]

We have the same problem here as in OpenCyphal/pycyphal#303 (review)

[clyde-johnston]

See OpenCyphal/pycyphal#303 for details.

[pavel-kirienko]

The exit status of nox is ignored on Windows; you can see that currently, the Windows job fails but the workflow status is green. Can you please check if this is also the case in the other projects?

I am not sure what the root case of the error is. It says that it is unable to bind a multicast socket to a local network interface (localhost in this case):

pycyphal.transport._error.InvalidMediaConfigurationError: Could not register multicast group membership 239.0.29.85 via 127.0.0.0 using <socket.socket [closed] fd=-1, family=AddressFamily.AF_INET, type=SocketKind.SOCK_DGRAM, proto=17> [WSAEADDRNOTAVAIL]

I am going to push an experimental change to collect more information about the problem.

[pavel-kirienko]

Yep, my speculative fix worked (not sure why it even worked on AppVeyor before the fix). Now the remaining issues are five timeout errors: https://github.com/OpenCyphal/yakut/actions/runs/5912398047/job/16035771126?pr=85#step:6:6988

One common cause for timeout errors on Windows that I often run into while testing is that the stderr buffer is very small, and if a child process generates log messages or runtime warnings in large amounts, it fills up the buffer and then at one point blocks forever on a write to stderr, which then triggers the test suite to report a timeout error. Looking into that now.