feat: 🎸 upgrade datasets to 2.8.0 (#666)
* feat: 🎸 upgrade datasets to 2.8.0
remove fix because https://github.com/huggingface/datasets/pull/5333 has
been included in the release, and remove deprecated use_auth_token
argument in download_and_prepare
* feat: 🎸 upgrade worker + enable parquet up to 5GB datasets
