Bump fsspec upper bound constraint

[hadim]

Would it be possible to bump fsspec upper bound to the latest (2026.2.0)?

I saw you had some API compat issues in the past (#7326) and I understand the need for an upper bound.

But I wonder if you think your CI and tests are a good proxy to catch fsspec API breakage? If that's the case then triggering dataset CI with the latest version of fsspec should tell us whether it's all good right?

Happy to open a PR if needed.