Fix multigpu test

[ksikka]

No description provided.

[themattinthehatt]

thanks for looking into this - any idea why it was working previously?

[ksikka]

Of course. I recently tested that multigpu tests pass in axon, then updated the srun to have gres:2 to allow multigpu tests to run in CI.

However sometime after that, I must have run pip install lightning-pose from pypi, not using -e, and this breaks the multigpu tests as the subprocess call to python script/train_hydra.py resolves import lightning_pose from the pip installed lightning pose, not the root of the git repo that you run pytest in. Pytest uses the prepend method described here, but it does not extend to subprocesses:
https://docs.pytest.org/en/7.1.x/explanation/pythonpath.html