Automatic test discovery sans boilerplate

[Helveg]

adresses #1687, closes #1463.

Removed all of the verbose deprecated boiler plate in favor of automatic discovery.

Threw in a little fixture system for extra points:

from .. import fixtures

@fixtures.context
@fixtures.empty_recipe
def test_something(context, empty_recipe):
  arbor.simulation(empty_recipe, context)

If anyone wants to add fixtures all they have to do is write a function/class in python/test/fixtures.py and decorate it with the internal _fixture or _singleton_fixture decorators. Then they can be used in test files as decorators that inject parameters with their name like in the example above:

@_fixture
class empty_recipe(arbor.recipe):
    pass

Fixtures added:

• @repo_path returns the pathlib.Path to the repo root
• @context returns an arbor.context with MPI (if enabled) and `mpi4py comm (if enabled). Auto inits MPI.
• @dummy_catalogue returns an arbor.catalogue with the dummy mech in it, uses build-catalogue, builds under MPI on one node, doesn't stall if there's build errors.
• @empty_recipe returns an instance of an arbor.recipe without anything defined in it.

[Helveg]

I require to build catalogues in CI, which overlaps with #1632, so I merged it in and will merge master once that PR is merged.

[Helveg]

I added cases.py as a centralized location for modifying TestCases and added some skip decorators that should replace the old seperate runs of unit and unit_distributed. Advantages are that unmarked tests are now run both with and without MPI and you can just run the suite and it will detect whether it can run MPI tests by itself, skipping appropriately:

# Skips all MPI only tests
python -m unittest discover -s python
# Skips all no-MPI only tests
mpiexec -n 2 python -m unittest discover -s python

Disadvantages are that tests that run exclusively with/without MPI need to be marked explicitly with the decorators, but that's 1 line of code per file.

[thorstenhater]

👍