Bumpy Numpy version limit to include 2.x only + Pytorch 2.6

[ericspod]

Partially addresses #8303.

Description

This changes the maximum Numpy version to be below 3.0 for testing with 2.x compatibility. This appears to be resolved with newer versions of dependencies. This will also include fixes for Pytorch 2.6 mostly relating to torch.load usage.

Types of changes

• Non-breaking change (fix or new feature that would not break existing functionality).
• Breaking change (fix or new feature that would cause existing functionality to change).
• New tests added to cover the changes.
• Integration tests passed locally by running ./runtests.sh -f -u --net --coverage.
• Quick tests passed locally by running ./runtests.sh --quick --unittests --disttests.
• In-line docstrings updated.
• Documentation updated, tested make html command in the docs/ folder.

[jamesobutler]

@ericspod I would suggest updating the automated workflow so that when it tests older torch versions (pre-2.3.0 which don’t have numpy 2 support), to python -m pip install numpy<2 just before the torch install linked below. This should keep this numpy 1 version as torch is installed and as well as when MONAI gets installed afterwards. With this PR’s change, having numpy<2 already installed with an older torch is fine and will meet the install_requires for MONAI when it goes to get installed and won’t force the upgrade of them.

MONAI/.github/workflows/pythonapp-min.yml

Line 157 in df1ba5d

python -m pip install torch==${{ matrix.pytorch-version }}

[ericspod]

@ericspod I would suggest updating the automated workflow so that when it tests older torch versions (pre-2.3.0 which don’t have numpy 2 support), to python -m pip install numpy<2

Thanks for the suggestion, that would fix this PR however we wouldn't resolve the version issue during installation. Users would still need to pre-install Numpy like this. This PR was more an experiment to see what version updates we can do and where the pain points were. We're still discussing the version policy which will change what we do here as well, and hopefully resolve this Numpy issue.