Do not use pathlib for hdf5 datasets, fixes #2318

[maxnoe]

fixes #2318

[Tobychev]

Is this use of paths inside of h5 files tested somewhere? I took a look at io/tests/test_hdf5.py but I nothing stands out as testing the access of paths inside of a file.

[kosack]

Was there a reason to implement this by hand rather than just switching to using PurePosixPath as stated in the issue? That should work, right? and it is a smaller change and leverages the standard library, rather than introducing more code. With that, split_hdf5_path can be removed altogether.

Checking for a leading '/' would then be done like this:

path = PurePosixPath(path)

if path.root != '/':
    raise ValueError("HDF5 paths should be absolute (start with '/') and not relative", path)

[maxnoe]

Was there a reason to implement this by hand rather than just switching to using PurePosixPath as stated in the issue?

I only used (and moved) the already existing functionality to fix this specific issue.

I agree that we probably should cleanup all our path handling in the hdf5 modules and PurePosixPath might be a good solution, but I'd keep that for another PR.