Set upper bound of PyTorch version

[fepegar]

Description

This doesn't work

import torch
import torchio as tio

subject = tio.datasets.Colin27()
subject.load()
subjects = 10 * [subject]

from torch.utils.data import DataLoader
loader = DataLoader(subjects, batch_size=4)
batch = next(iter(loader))
batch.__class__ is dict

because batch is an instance of Subject.

This happens in PyTorch >=2.3 because of

• Preserve metadata for MutableMapping and MutableSequence in pin_memory and collate_fn pytorch/pytorch#120553

This PR forces PyTorch to be <2.3, temporarily.

Checklist

• I have read the CONTRIBUTING docs and have a developer setup (especially important are pre-commitand pytest)
• Non-breaking change (would not break existing functionality)
• Breaking change (would cause existing functionality to change)
• Tests added or modified to cover the changes
• Integration tests passed locally by running pytest
• In-line docstrings updated
• Documentation updated, tested running make html inside the docs/ folder
• This pull request is ready to be reviewed

[fepegar]

@justusschock, do you think you could take a look at this problem? As you worked on the logic for copying subjects. If not, that's absolutely fine, but please let me know.

[fzimmermann89]

@fepegar : Maybe a stupid question, but why is this important?

isinstance(batch,dict) is still true, as Subject is a subclass of dict.
This seems quite logical and works as I would expect.

(I am asking because we would like to keep using torchio with pytorch 2.3 and would be willing to fix the issue -- if I could understand why this should be an issue and what the expected behavior is)