upload datasets to huggingface

[kjappelbaum]

• if the renaming is not too tricky, I'd consider renaming the representations to align with the paper
• on huggingface, we probably want to have one repository and then a dropdown ("subsets") for the different pertaining sizes

[kjappelbaum]

then also ensure we have https://github.com/mlcommons/croissant for the checklist

[kjappelbaum]

include in README instructions to download the data

[kjappelbaum]

ok, the remaining thing here is to update the docs