Parallelize Python test suite with pytest-xdist (31s -> 12s)

[Aztec03hub]

What

Parallelize the Python unit-test suite with pytest-xdist. The 1526-test suite was running serially on a single core (~63% CPU on an 8-core box). This defaults the run to -n auto --dist=worksteal, a pure wall-clock win with identical results.

Results (8-core box)

	wall clock	test time	CPU
Before (serial)	~31s	~26s	61%
After (-n auto)	~12s cold, ~6-9s warm	~9s	307%

~2.6x faster cold wall-clock, same 1526 passed, zero skips added.

Changes

• Add pytest-xdist>=3.0 to the dev optional-dependency group.
• Add addopts = "-n auto --dist=worksteal" to [tool.pytest.ini_options] (existing testpaths and filterwarnings untouched). worksteal rebalances pending tests onto idle workers, which matters because the textual TUI tests are much slower than the rest.

Parallel safety

No production or test logic needed changing — the suite was already parallel-safe, verified by audit:

• Real socket binding (test_mqtt_ws_bridge.py, test_doctor.py) already uses OS-assigned ephemeral ports via socket.bind((host, 0)) / getsockname().
• Broker / publisher lifecycle tests (test_broker.py, test_publisher.py) fully mock amqtt / aiomqtt, so no real ports are opened.
• Filesystem / env / home / cwd state flows through per-test tmp_path, monkeypatch.setenv, and patched Path.home(). Each xdist worker is its own process with its own tmp base, so 8 workers never collide. No test writes to a real ~/.claude-comms or a repo-relative path.
• The fixed ports (1883 / 9001 / 9920 / 9921) in conftest's tmp_config / e2e_config are config-dict values only and are never bound to a socket.

CI

The Test job runs plain pytest --tb=short --junitxml=... with no addopts override and no -p no:xdist, so it inherits -n auto automatically. pytest-xdist ships via the [all,dev] extra the job already installs. Verified the exact CI command works under xdist (junitxml records tests="1526", exit 0).

Flake gate — 5 consecutive parallel runs

Run 1: exit=0 | 1526 passed in 6.31s
Run 2: exit=0 | 1526 passed in 6.67s
Run 3: exit=0 | 1526 passed in 8.21s
Run 4: exit=0 | 1526 passed in 8.64s
Run 5: exit=0 | 1526 passed in 9.75s

All 1526 passed, all exit 0 — no parallel flakiness. ruff check, ruff format --check, and basedpyright (src 0/0, tests 0/0) all remain clean.

🤖 Generated with Claude Code