Fix OSCAR Esperanto

[lhoestq]

The Esperanto part (original) of OSCAR has the wrong number of examples:

from datasets import load_dataset
raw_datasets = load_dataset("oscar", "unshuffled_original_eo")

raises

NonMatchingSplitsSizesError:
[{'expected': SplitInfo(name='train', num_bytes=314188336, num_examples=121171, dataset_name='oscar'),
'recorded': SplitInfo(name='train', num_bytes=314064514, num_examples=121168, dataset_name='oscar')}]

I updated the number of expected examples in dataset_infos.json

cc @sgugger