`datasets` no longer supports dataset loading scripts, so `load_dataset('biglam/artigo')` fails with `RuntimeError: Dataset scripts are no longer supported`.
This PR converts the dataset to parquet and removes the loading script. The schema is unchanged, including the nested per-tag `regions`. Images are embedded, so loading no longer depends on Zenodo.
(The script would also fail today for a second reason: it selected the Zenodo files on `file['type']`, a key the current Zenodo API no longer returns.)
Source data unchanged: https://doi.org/10.5281/zenodo.8202331 (CC BY-SA 4.0).
