userspace socket: treat a closed peer handle as writable

[pianiststickman]

Commit Message: userspace socket: treat a closed peer handle as writable
Additional Description:
If one side of an internal listener connection closes, subsequent writes from the other side may result in data waiting to be flushed on a half-close. For connections managed by the HCM, this is covered by the delayed close timer, but TcpProxy does not implement a delayed close timer and therefore the connection leak in #38148 can persist.

The fix here is treat a closed peer handle as writable in order to get the connection to attempt a write and error out. (This is analogous to Linux signalling EPOLLOUT along with EPOLLHUP when a peer closes for reads; without it, an epoll consumer on the other side may continue to write-block a socket indefinitely, which is exactly what's happening here but in userspace.)
Risk Level: low; subsequent calls to write() are guarded against peer being nullptr, and the error when calling write() is the point.
Testing: new CI and integration test that reproduces the issue
Docs Changes: n/a
Release Notes: n/a
Fixes #38148

[repokitteh-read-only]

As a reminder, PRs marked as draft will not be automatically assigned reviewers,
or be handled by maintainer-oncall triage.

Please mark your PR as ready when you want it to be reviewed!

🐱

Caused by: #43976 was opened by pianiststickman.

see: more, trace.