NumPy 2.0 support

[rgommers]

NumPy 2.0 is coming out soon ( numpy/numpy#24300 ). NumPy 2.0.0rc1 packages for conda & wheels came out 6 weeks ago ( numpy/numpy#24300 (comment) ).

To prepare for NumPy 2.0, it might be worthwhile to start testing against NumPy 2 in CI

NumPy has put out a migration guide. More details are in the release notes

Clawpack does make use of NumPy's C API (via f2py) and produces one wheel per release (for macOS arm64, at https://pypi.org/project/Clawpack/5.10.0/#files), which will break. Therefore, having a release of XGBoost with wheels built against NumPy 2.0.0rc1 would be helpful to ensure that wheel is NumPy 1 & 2 compatibl (as wheels built against NumPy 1 won't be compatible with NumPy 2). More details in this NumPy 2 ABI doc

There are a number of C API usages in clawpack/pyclaw#712 that currently break the build against NumPy 2.0 RCs, so that PR should be merged. There may be other issues (I haven't checked).

Note that there is currently no cap on the numpy version in pyproject.toml:

clawpack/pyproject.toml

Line 50 in 4c5990f

"numpy",

That means that building from source with for example pip install clawpack will also break when NumPy 2.0 is released. Doing a new release with either numpy 2.0 support or at least a <2 version cap seems advisable.