clarification required for the validation split

it is unfortunately not clear how the validation split is defined for each dataset. As the list of fields only indicate 'test_split'. Are we to find this information on OpenML website? Also, there appears no information on how to recreate the train/val/test splits.

Review how the benchmark currently represents each dataset's fields, especially the test_split value, and check the corresponding OpenML information. Document where the validation split is defined and how the train, validation, and test splits can be recreated, then verify the explanation against the dataset records.

Written by the indexing model from the issue text.