bpo-30064: Fix asyncio loop.sock_* race condition issue

[fantix]

https://bugs.python.org/issue30064

This is reproducible by a cancel() followed directly by a retry:

done, pending = await asyncio.wait([loop.sock_recv(sock, 1024)], timeout=0.1)
if done:
    data = await done.pop()
else:
    pending.pop().cancel()
    data = await loop.sock_recv(sock, 1024)

This issue applies to all the loop.sock_* methods. Without this PR, users could work it around by adding a sleep(0):

done, pending = await asyncio.wait([loop.sock_recv(sock, 1024)], timeout=0.1)
if done:
    data = await done.pop()
else:
    pending.pop().cancel()
    await asyncio.sleep(0)  # <---- this ensures that the right reader is removed
    data = await loop.sock_recv(sock, 1024)

[1st1]

@asvetlov Please take a look if you have a few minutes.

[miss-islington]

Thanks @fantix for the PR, and @1st1 for merging it 🌮🎉.. I'm working now to backport this PR to: 3.9.
🐍🍒⛏🤖

[1st1]

Thanks, Fantix!

[bedevere-bot]

GH-20460 is a backport of this pull request to the 3.9 branch.