fix: moran's calculation parallelization

[ilan-gold]

Was getting the following beginning with numba>=0.66rc1:

Details

=================================== FAILURES =================================== __________________ test_consistency[morans_i-single-threaded] __________________ [gw1] linux -- Python 3.14.5 /home/runner/.local/share/hatch/env/virtual/scanpy/B9PcT7QG/hatch-test.pre/bin/python3

metric = <function morans_i at 0x7fae9eda6140>

@pytest.mark.usefixtures("_threading")
def test_consistency(metric) -> None:
    pbmc = pbmc68k_reduced()
    pbmc.layers["raw"] = pbmc.raw.X.copy()
    g = pbmc.obsp["connectivities"]
    equality_check = partial(np.testing.assert_allclose, atol=1e-11)

    # This can fail
    equality_check(
        metric(g, pbmc.obs["percent_mito"]),
        metric(g, pbmc.obs["percent_mito"]),
    )
    equality_check(
        metric(g, pbmc.obs["percent_mito"]),
        metric(pbmc, vals=pbmc.obs["percent_mito"]),
    )

    equality_check(  # Test that series and vectors return same value
        metric(g, pbmc.obs["percent_mito"]),
        metric(g, pbmc.obs["percent_mito"].values),
    )

    equality_check(
        metric(pbmc, obsm="X_pca"),
        metric(g, pbmc.obsm["X_pca"].T),
    )

    all_genes = metric(pbmc, layer="raw")
    first_gene = metric(
        pbmc, vals=pbmc[:, pbmc.var_names[0]].layers["raw"].toarray().ravel()
    )

  np.testing.assert_allclose(all_genes[0], first_gene, rtol=1e-9)

E AssertionError:
E Not equal to tolerance rtol=1e-09, atol=0
E
E Mismatched elements: 1 / 1 (100%)
E Max absolute difference among violations: 7.9750179849e-08
E Max relative difference among violations: 1.8702209022e-07
E ACTUAL: array(0.426421)
E DESIRED: array(0.426421)

tests/test_metrics.py:72: AssertionError
__________________ test_consistency[morans_i-multi-threaded] ___________________
[gw1] linux -- Python 3.14.5 /home/runner/.local/share/hatch/env/virtual/scanpy/B9PcT7QG/hatch-test.pre/bin/python3
data[i] -= regressor[i] @ coeff

tests/test_scaling.py::test_scale[mask-center-int64-array-csc_matrix]
tests/test_scaling.py::test_scale[mask-center-int64-anndata-csc_matrix]
tests/test_scaling.py::test_scale[mask-center-float32-anndata-csc_matrix]
tests/test_scaling.py::test_scale[mask-center-float32-array-csc_matrix]
/home/runner/.local/share/hatch/env/virtual/scanpy/B9PcT7QG/hatch-test.pre/lib/python3.14/site-packages/scipy/sparse/_index.py:216: SparseEfficiencyWarning: Changing the sparsity structure of a csc_matrix is expensive. lil and dok are more efficient.
self._set_arrayXarray(i, j, x)

tests/test_scaling.py::test_scale[mask-center-int64-anndata-csr_matrix]
tests/test_scaling.py::test_scale[mask-center-float32-array-csr_matrix]
tests/test_scaling.py::test_scale[mask-center-float32-anndata-csr_matrix]
tests/test_scaling.py::test_scale[mask-center-int64-array-csr_matrix]
/home/runner/.local/share/hatch/env/virtual/scanpy/B9PcT7QG/hatch-test.pre/lib/python3.14/site-packages/scipy/sparse/_index.py:216: SparseEfficiencyWarning: Changing the sparsity structure of a csr_matrix is expensive. lil and dok are more efficient.
self._set_arrayXarray(i, j, x)

-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html

• generated xml file: /home/runner/work/scanpy/scanpy/test-data/test-results.xml -
  =========================== short test summary info ============================
  FAILED tests/test_metrics.py::test_consistency[morans_i-single-threaded] - AssertionError:
  Not equal to tolerance rtol=1e-09, atol=0

Mismatched elements: 1 / 1 (100%)
Max absolute difference among violations: 7.9750179849e-08
Max relative difference among violations: 1.8702209022e-07
ACTUAL: array(0.426421)
DESIRED: array(0.426421)
FAILED tests/test_metrics.py::test_consistency[morans_i-multi-threaded] - AssertionError:
Not equal to tolerance rtol=1e-09, atol=0

Mismatched elements: 1 / 1 (100%)
Max absolute difference among violations: 7.9750179849e-08
Max relative difference among violations: 1.8702209022e-07
ACTUAL: array(0.426421)
DESIRED: array(0.426421)

and suspect this could be the cause

• Closes #
• Tests included or not required because:

• Release notes not necessary because: fix in preparation for upcoming release

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 79.61%. Comparing base (d66537f) to head (073472b).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #4148      +/-   ##
==========================================
+ Coverage   79.59%   79.61%   +0.02%     
==========================================
  Files         120      120              
  Lines       12786    12786              
==========================================
+ Hits        10177    10180       +3     
+ Misses       2609     2606       -3     

Flag	Coverage Δ
hatch-test.low-vers	78.83% <ø> (-0.01%)	⬇️
hatch-test.pre	79.48% <ø> (+0.03%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
src/scanpy/metrics/_morans_i.py	100.00% <ø> (ø)

... and 1 file with indirect coverage changes

[ilan-gold]

At this point, I think this may just be worth reporting upstream to numba and loosening our tolerance