perf: refactor to lazy interface loading

[roelofvandijk]

Relates to #1184

By moving the car_helpers.py imports to a pytest.fixture, we can shorten the collection time.

Refactors the car tests to use:

• a pytest fixture for the interface loading (this import is not executed during collection)
• pytest.mark.parametrize instead of the parametrize in select places to enable properly using the fixture. coincidentally already fixed in Replace @parameterized.expand with @pytest.mark.parametrize to improve test speed #1930

I tried to keep the diff as small as possible.

There are 3 big chunks of time spent during the pytest collection on a fresh run (after compilation):

• first import of CANDefine, aka the packer_pyx compiled library (only slow the first run after compilation)
• import of the interfaces in car_helpers.py (always slow)
• import of pytest plugins

"Fresh" denotes a recompilation of the cython modules.

Collect only

pytest opendbc/car/tests --collect-only

State	Fresh	Second Run
master	0.56s	0.27s
PR	0.16s	0.13s

compare to

pytest opendbc/can/tests --collect-only

State	Fresh	Second Run
master	0.32s	0.02s
PR	0.29s	0.04s

Collection + execution

pytest opendbc/car/tests

State	Fresh	Second Run
master	6.42s	6.19s
PR	5.92s	5.76s

[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

[sshane]

Can we do this without changing the structure of this test too much?

[roelofvandijk]

If this is to make the review easier, yes. I reduced the diff.

It seems nonsensical, because we are not parametrizing anymore over the test class, but directly over the test functions.