Bump datasets version

[JanetVictorious]

Add support for datasets>2.17. This PR resolves issue #1136.

Change is to support version 3.0.0 and upwards.

[github-actions]

DCO Assistant Lite bot All contributors have signed the DCO ✍️ ✅

[JanetVictorious]

I have read the DCO Document and I hereby sign the DCO

[JanetVictorious]

recheck

[JanetVictorious]

Can this be merged or are there any changes that need to be made? Ping @leondz @jmartin-tech

[leondz]

Will take a look (alle i virksomheden verden over havde ferie de sidste to dage i sidste uge :) )

[leondz]

• we looked at this last week: current version pinning in garak is due to an issue w/ some versions 2.17.x - see Bump datasets package #536 and also upstream datasets bug Error loading wikitext data raise NotImplementedError(f"Loading a dataset cached in a {type(self._fs).__name__} is not supported.") huggingface/datasets#6352 (comment). Working out a test for this issue so we can safely bump versions in future. Once a relevant test is in place this PR can progress.

[JanetVictorious]

Ok thanks for the update and the links, I will hold off then until it's verified we can safely bump the the version and then maybe skip 2.18.0 if it also entails problems (härligt med lite semester :) )

[jmartin-tech]

The original restriction here was put in place due to a transitive dependency on fsspec[http] that still holds the same minimum version restriction in datasets though some mitigation code has been added that should ensure compatible function with old and newer versions of fsspec. While basic testing shows this works with datasets>=2.18.0 it may be more appropriate to bump this to datasets>=3.0.0 updating to a expect compatibility with the latest major version.