automated combined public dataset

we currently have a set of public datasets, either as splits or entirely private.
It could be quite impactful for the community to automatically and periodically combine these into a large dataset release, e.g. for model training.

The issue does not name files, tests, or an existing release entry point. Start by locating how public datasets and dataset splits are defined and released, then determine how periodic combination and access should work. Done means an agreed automated process can produce a large public dataset release for model training.

Written by the indexing model from the issue text.