Improve test coverage for core modules using shared fixtures

[jwrth]

Background

Overall test coverage is ~33% (4,525 / 13,801 statements). The largest untested modules were assessed in March 2026. Two modules have 0% coverage and are good near-term targets. Two megaclass modules are feasible but require shared fixture infrastructure first.

Priority targets

Module	Statements	Coverage	Barrier
tools/permutation.py	236	0%	Pure numerical logic; needs AnnData + spatial graph fixture
spatialdata/_convert.py	261	0%	Conversion layer; needs round-trip InSituPy ↔ SpatialData fixture
_core/data.py	731	26%	Core InSituData class; needs shared fixture
experiment/data.py	947	14%	Multi-sample container; needs shared fixture

Skip for now: interactive/_widgets.py and interactive/_transcript_viewer.py — both require a live napari/Qt display context.

What needs to be done

1. Build a shared fixture in tests/conftest.py — a minimal but real InSituData object (and optionally a small InSituExperiment) constructed once per session:

   @pytest.fixture(scope="session")
   def insitudata_small():
       # load smallest available test dataset or construct synthetically
       ...

2. Cover permutation.py — test the permutation null distribution logic with a small AnnData + precomputed neighbor graph.
3. Cover spatialdata/_convert.py — test round-trip conversion (InSituPy → SpatialData → InSituPy) for each element type (images, tables, annotations, transcripts, boundaries).
4. Cover _core/data.py and experiment/data.py using the shared fixture once it exists.

Implementation prompt

You are adding tests to the InSituPy package to improve coverage of four under-tested modules. Start by creating a shared session-scoped pytest fixture in tests/conftest.py that constructs a minimal InSituData object (use the smallest available test dataset or synthetic data). Use the conda environment isp012 to run pytest.

Then implement tests in this order:

1. tests/test_tools_permutation.py — cover tools/permutation.py (236 stmts, 0%). Build a small AnnData with a precomputed neighbor graph and test the permutation null distribution logic.
2. tests/test_spatialdata_convert.py — cover spatialdata/_convert.py (261 stmts, 0%). Write round-trip conversion tests (InSituPy → SpatialData → InSituPy) for each supported element type.
3. tests/test_core_data.py — cover _core/data.py (731 stmts, 26%) using the shared insitudata_small fixture.
4. tests/test_experiment_data.py — cover experiment/data.py (947 stmts, 14%) using a shared InSituExperiment fixture wrapping multiple InSituData instances.

Run pytest --cov=insitupy --cov-report=json before and after to measure the coverage delta. Target: raise overall coverage above 45%.