Standardize the inputs for dataset-ingestion.

[gsaluja9]

At the moment, there are 2 datasets that are available to ingest in ds-ingest workflow.
They have scripts to process the actual dataset, as the author published, and convert them to input csvs for aperturedb.

They have been published in variety of ways (even between datasets, eg. CelebA vs CelebA-hq), so their intermediate processing is handled in a bespoke manner.

The ask here is that we convert these datasets to something like a csv or croissant so that Dataset-ingestion is simple and easier to mantain.

The cons to maintaining this kind of input is that if the source dataset changes, our intermediate representation would need to be refreshed, however this might not be such a big deal as most of the datasets (like COCO), come with a version associated, and are built incrementally.