Fix SocketStream.aclose() abort() race on the asyncio backend

[afonsojanu]

NOTE Erasing or replacing the contents of this template will result in your pull
request being summarily closed without consideration!

Changes

Fixes #1250.

On the asyncio backend, SocketStream.aclose() closes the transport and then
checkpoints with await sleep(0) before calling transport.abort(). If the
transport's write buffer drains and connection_lost() fires during that
checkpoint, the transport has already detached itself from the event loop by
the time aclose() resumes, so the subsequent abort() call raises
AttributeError: 'NoneType' object has no attribute 'call_soon' instead of
being the harmless no-op it's meant to be in that interleaving.

StreamProtocol now records whether connection_lost() has already run, and
aclose() skips the abort() call when it has, since the connection is
already gone at that point.

Added a deterministic regression test that schedules connection_lost() to
fire during the checkpoint (via loop.call_soon) rather than relying on the
timing-dependent socket-buffer repro from the issue.

Checklist

• You've added tests (in tests/) which would fail without your patch
• You've updated the documentation (in docs/), in case of behavior changes or new
  features
• You've added a new changelog entry (in docs/versionhistory.rst).

[agronholm]

There are now tons of PRs on this same subject, and I'm drowning in them. Justify why you had to send yet another one.

[afonsojanu]

Fair question. I went back and checked, and #1255 already covers this exact bug (same issue #1250, same fix approach: track connection-lost state on the protocol and skip abort() when it's already true). It's been open since July 30th, so it predates mine by a month.

I should have searched open PRs against issue #1250 before starting instead of just checking for an existing fix on master. Closing this one, sorry for adding to the pile.