Documenting existence of cross-validation folds

[hayesall]

Currently, there's no indication in the documentation for which datasets have standard cross-validation folds, or which are simply split into train / test.

---

Alternatively, it might be useful to think of a metadata attribute on the datasets:

from relational_datasets import get_metadata

get_metadata("toy_cancer")
{
  'name': 'toy_cancer',
  'folds': 1,
  'modes': ['smokes(+person).', 'cancer(+person)', ...],
}

Possibly related to: srlearn/datasets#11