Make tests run without Internet usage

[noahho]

Describe the workflow you want to enable

tests are using datasets from sklearn / openml that need to be downlaoded first. This sometimes crashes tests.

Describe your proposed solution

Pre-load datasets

Describe alternatives you've considered, if relevant

No response

Additional context

No response

Impact

None

[linear[bot]]

from oscar:

After https://github.com/PriorLabs/TabPFN/pull/481 we just have the iris dataset in the consistency tests remaining.