`subint_forkserver`: `test_nested_multierrors` cancel-cascade hang gated by pytest `--capture=fd`

[goodboy]

tests/test_cancellation.py::test_nested_multierrors[subint_forkserver]
hangs the full multi-level (root → spawner → grandchild → errorer)
cancel cascade under --spawn-backend=subint_forkserver. Several
fixes landed during PR #447 closed earlier layers of the bug — FD
hygiene via _close_inherited_fds, pidfd_open-based
_ForkedProc.wait, _parent_chan_cs shielded-loop wiring, bounded
wait_for_no_more_peers — but the final-gate behavior was
empirically traced to pytest's own --capture=fd stdout/stderr pipe
filling under the high error-traceback volume of the test; only
pytest -s makes the test pass cleanly. The skip-mark
skipon_spawn_backend('subint_forkserver') remains on the test.

Open follow-ups

• Decide whether to redirect subactor stdout/stderr to /dev/null
  (or a session-scoped log file) in the fork-child prelude, OR
  document pytest -s (or --capture=sys) as required for
  multi-level forkserver-backend tests.
• Identify whether any unbounded wait remains above async_main in
  the teardown cascade once the capture-pipe gate is bypassed.

See also

ai/conc-anal/subint_forkserver_test_cancellation_leak_issue.md
— full multi-session post-mortem with the cascade-tree view and the
"VERY late: pytest capture pipe is the final gate" finding.

---

Tracked from #379 (subint umbrella).