Remove NumPy <2 pin

[seberg]

This PR removes the NumPy<2 pin which is expected to work for
RAPIDS projects once CuPy 13.3.0 is released (CuPy 13.2.0 had
some issues preventing the use with NumPy 2).

[jakirkham]

Now that CuPy 13.3.0 is out. Marking this ready for review

Merged in branch-24.10 latest to pickup the fixes in PR: #764

[jakirkham]

Looks like the wheels had two test failures in CI:

FAILED python/cucim/src/cucim/skimage/color/tests/test_colorconv.py::test_gray2rgba_alpha_fail_cast[-1-uint8] - OverflowError: Python integer -1 out of bounds for uint8
FAILED python/cucim/src/cucim/skimage/color/tests/test_colorconv.py::test_gray2rgba_alpha_fail_cast[300-int8] - OverflowError: Python integer 300 out of bounds for int8

Conda doesn't have the same issue, but am seeing NumPy downgraded to there. So we are likely missing something there too

[jakirkham]

So Conda downgrades NumPy because

1. Conda-forge rebuilt scikit-image 0.24.0 with NumPy 2: Rebuild for numpy 2.0 conda-forge/scikit-image-feedstock#110
2. Currently we only support scikit-image 0.23.x: Upgrading to scikit-image 0.24.0 #768

Addressing either of these would fix the issue

[seberg]

So this failure is new with NumPy 2.1, it seems pretty harmless, I pushed a potential fix, but would be good if @grlee77 can have a quick look if that seems good.

[jameslamb]

Packaging changes look ok to me. I poked around and don't see any other uses of numpy that were missed.

Just one question... have we double-checked that the assumptions here hold for both the NumPy 1.x and 2.x ABI?

cucim/cpp/include/cucim/memory/dlpack.h

Line 33 in de6bfbb

inline const char* to_numpy_dtype(const DLDataType& dtype) {

Are there new-in-NumPy-2.x datatypes that need to be added there?

[grlee77]

Thanks, @seberg. The compatibility change looks good to me

[jakirkham]

Thanks all! 🙏

The changes look good here

We do need to update scikit-image, which is being done in PR: #769

So marking as requests changes and adding "DO NOT MERGE" until that PR is in

Once that PR is in will merge in the latest upstream branch here to rerun CI and confirm that we have fixed the Conda NumPy 1 downgrade issue mentioned above. Then validate the CI logs to make sure we are getting NumPy 2

Assuming all of that goes well. We can merge after

[jakirkham]

PR ( #769 ) was just merged

Have resolved conflicts (as numpy & scikit-image dependencies were on adjacent lines in come cases)

CI is now running

[jakirkham]

Ok can now see NumPy 2 in CI testing of conda packages and it remains through later updates

numpy                     2.1.0           py311hed25524_0    conda-forge

Approving now that this issue is cleared up

[jakirkham]

/merge

[jakirkham]

Thanks all! 🙏