mazda: fix the CX-9 model tests and run the fork's tests in CI

[zephleggett]

Three failures that predate #8 and fail on every branch.

test models: MAZDA_CX9 radar and canValid

MazdaPlatformConfig's default dbc_dict gained Bus.radar for the CX-5 2022 track work, so every Mazda claimed a radar and radarUnavailable was False everywhere. The 2016-20 CX-9 is the one platform that never puts the 0x361-0x366 tracks on bus 0, so RadarInterface built a parser that never went valid.

Checked all six Mazda platforms against segments 0-2: only the CX-9 lacks the tracks, so it gets its own dbc_dict without the radar bus.

The can_invalid_cnt failure was separate and only reproduced on segment 0, which is the segment CI happens to pick. CAM_PEDESTRIAN is read through vl, so it registered lazily with liveness checks on while its three siblings were already declared at nan for exactly that reason. The 2016-20 CX-9 camera never sends it, so canValid was false for the whole route (5745 frames).

./test.sh: the fork's tests never ran

opendbc has no pytest anywhere - the suite runs under unittest-parallel, safety under unittest discover, and pytest is in neither pyproject.toml nor uv.lock. These five files were the only ones importing it, so discovery raised ModuleNotFoundError on each. The 146 tests have never executed in this repo; they only ran in the parent openpilot repo, whose venv has pytest.

Converted in place rather than adding pytest, which would put a dependency upstream does not have into files upstream owns. Fixtures become module-level factories, parametrize becomes a class-level case table walked with subTest, and approx becomes assertAlmostEqual or a small _Approx helper where the tolerance is compared inside an expression. Case coverage is unchanged at 166.

TestStandstillHold and TestAdvertisedLead had a run() helper that shadows TestCase.run; it is drive() now.

Discovery goes from 9529 tests with 5 errors to 9656 with none.

Not addressed

The Safety mutation tests failure is flaky, not a real regression: it passed on 8ec52e0 and failed on f8c21af, and the only diff between them is a comment rewrite. The four survivors are in lateral.h and longitudinal.h, which #8 never touched.

Verified locally

• all six Mazda platforms, test_car_interface + test_radar_interface, segments 0-2
• unittest-parallel -j4: 9656 tests, OK
• ruff and codespell clean

[github-actions]

Thanks for contributing to opendbc! In order for us to review your PR as quickly as possible, check the following:

• Convert your PR to a draft unless it's ready to review
• Read the contributing docs
• Before marking as "ready for review", ensure:
  • the goal is clearly stated in the description
  • all the tests are passing
  • include a route or your device' dongle ID if relevant