Use `conftest.py` for shared fixtures

[contsili]

Our tests currently rely on __init__.py files so that shared fixtures can be imported explicitly, e.g. from test.fixtures.<module> import .... This makes test/ an importable package.

scientific-python recommends against putting __init__.py in the tests directory (as it can confuse package-discovery), and share fixtures through conftest.py instead, which pytest auto-discovers so no import is needed (yay!). See https://learn.scientific-python.org/development/guides/pytest/

Note: conftest.py only auto-shares @pytest.fixture-decorated objects. Plain helper functions, stub model classes, and constants currently living in the fixture modules are not auto-imported.

Proposed change

1. Remove __init__.py from the test directories.
2. Move shared fixtures into conftest.py files, scoped by location:
   • Fixtures used across all tests → top-level test/conftest.py.
   • Fixtures specific to a subtree → a local conftest.py, e.g. test/test_longitudinal/conftest.py.
3. Replace explicit fixture imports (from test.fixtures... import ...) with fixtures injected by argument name.

[contsili]

Principles we should follow

• in our toolbox fixtures can be data, a pre configured model or a flexibly configurable model (see

  PCNtoolkit/test/fixtures/blr_model_fixtures.py

  Line 44 in 50f58f1

  def blr_model_factory() -> Callable:

  ) , a fitted model.

• About data: Test data often needs to vary per test (number of subjects, covariates, batch effects) -> either have flexibly configurable data or just create new data in each test.

• When the same test should hold for several configurations, use @pytest.mark.parametrize

• Prefer plain test functions with fixtures (pytest standard) over classes with self grouping (unittest standard) (it is ok though to keep the classes we already have in this test:

  PCNtoolkit/test/test_core/test_normative_model_main.py

  Line 11 in 50f58f1

  class TestNormativeModel:

  ).

• make sure tests are reproducible via seeding (rng)

• make warnings actually fail the test. If we allow a warning then use ignore it per test via @pytest.mark.filterwarnings("ignore:...")