ci: run the test suite on free-threaded CPython (3.14t)

[bagowix]

Summary

interlock has always documented the breaker as thread-safe: one threading.Lock
in _engine.py guards admission and recording, and the protected callable runs
outside it. Every interpreter in the CI matrix had a GIL, so nothing in CI could
falsify that claim.

3.14t (free-threaded) joins the quality matrix as a required job, and a
new tests/test_concurrency.py drives a single breaker from many threads at
once, asserting what the lock exists to provide:

• window counts add up under concurrent recording;
• snapshot() never returns a torn view;
• the HALF_OPEN caps (max_concurrent_probes, permitted_calls_in_half_open)
  are never exceeded;
• a Registry hands out exactly one breaker per name;
• a concurrent trip emits exactly one CLOSED → OPEN.

No races were found in the code as it stands. The tests are sensitive rather
than decorative — with Engine._lock replaced by a nullcontext, two of them
fail on 3.14t (115 of 192 records lost; the transition emitted twice) while all
five still pass on 3.11. That gap is the whole argument for the new job.

fakeredis is now imported only when it is actually the backend. Its Lua engine
(lupa) has not declared free-threading support, so importing it on 3.14t
re-enables the GIL with a RuntimeWarning — and filterwarnings = "error" turns
that into a collection error for the entire file. CI already points the job at a
real Redis service container, so the extra import was never needed there.

Deviation from the issue

The issue proposed pytest-run-parallel to re-run an existing subset
(test_breaker.py, test_registry.py, test_engine.py, test_coordination.py,
test_observability.py) from N threads. I tried it and dropped it: those tests
assert exact per-instance event counts, so running each of them from 4 threads
produces ~100 failures that are test-design artefacts (assert 10 == 5), plus
hard errors where it meets pytest-asyncio (a coroutine was expected, got None). It would have added a dev dependency that could only be made green by
weakening the assertions it runs. The acceptance criteria it was meant to serve
are met by test_concurrency.py, which owns its threads and can therefore
measure peak concurrency and rendezvous deterministically.

Checklist

• Tests added or updated (suite stays at 100% coverage)
• uv run ruff format --check and uv run ruff check pass
• uv run mypy and uv run pyright pass
• Docs updated (docs/) for user-facing changes
• CHANGELOG.md [Unreleased] updated
• Commits follow Conventional Commits

Verified locally on cpython-3.14.5+freethreaded: 523 passed (suite minus the
Redis file, which needs a server there), 100% coverage on 3.11, and the
contention tests green 15/15 consecutive runs on 3.14t.

Related issues

Closes #102

[github-actions]

Coverage report

This PR does not seem to contain any modification to coverable code.

[codspeed-hq]

Merging this PR will not alter performance

✅ 21 untouched benchmarks

---

_{Comparing feat/free-threaded-ci (8b2f171) with main (19e5460)}