This reverts commit 44b83210c8ad ("unix: Fix a socket leak").
That commit fixed the leak by holding the vnode pool lock until the new
socket was enqueued, but that meant threading a mutex and a vnode through
unp_connectat_peer() and unp_vnode_peer(), and it only covered
pathname peers. A peer named by a descriptor (an empty path, or
/dev/fd/N) is only soref'd, since the file reference is dropped
before connecting, so a racing close() could still drain the queue
under us. @markj and I both found the fix unfortunate.
D61078 closes the race in solisten_enqueue() itself, for every way of
reaching the peer, so the mutex and vnode plumbing is no longer needed
and can go.
Depends on D61078
Assisted-by: Claude Code (Opus 5)