Remove NumPy < 2.2 version pin for Python 3.14 once upstream bug is fixed

[trissim]

Issue

We currently pin NumPy to < 2.2 for Python 3.14 due to a compatibility bug in NumPy 2.2.x that causes ValueError: operands could not be broadcast together with shapes in scikit-image's histogram/bincount functions.

Error Details

ValueError: operands could not be broadcast together with shapes (256,) (258,) (256,)
  File "skimage/filters/thresholding.py", line 390, in threshold_otsu
    counts, bin_centers = _validate_image_histogram(image, hist, nbins)
  File "skimage/exposure/exposure.py", line 274, in histogram
    hist, bin_centers = _histogram(image, nbins, source_range, normalize)
  File "numpy/lib/_histograms_impl.py", line 876, in histogram
    n += np.bincount(indices, weights=tmp_w, minlength=n_equal_bins).astype(ntype)

Current Workaround

In pyproject.toml:

"numpy>=1.26.4,<2.2; python_version>='3.14'",
"numpy>=1.26.4; python_version<'3.14'",

Action Items

• Monitor NumPy releases for fix (check NumPy 2.2.1+)
• Monitor scikit-image releases for workaround
• Test with newer NumPy versions when available
• Remove version constraint once bug is fixed
• Update CI to test with latest NumPy

References

• First observed in CI: PR Fix Windows virtual workspace path separator issue #33
• Affects: Python 3.14 only
• Workaround implemented: commit 00b9a4e

How to Test

Once a fix is available:

# Remove the version pin from pyproject.toml
# Then test:
python3.14 -m pytest tests/integration/test_main.py \
  --it-backends disk \
  --it-microscopes ImageXpress \
  -v

If tests pass, the pin can be removed.