cache public datasets and use them in integration tests

[dkrako]

Description of the problem

There were several cases (and #517 is still open) during this project, where going through a complete 10 minutes pymovements tutorial wouldn't have worked for the actual public datasets, besides the toydataset used in the jupyter notebook.

Up to now we ususally decided against testing these public datasets, as downloading all the datasets for each and every commit here on github would be a total waste of resources.

Description of a solution

But there is a github action for caching dependencies: https://docs.github.com/en/actions/using-workflows/caching-dependencies-to-speed-up-workflows#comparing-artifacts-and-dependency-caching

It seems that github workers could then reuse resuources for 90 days, which could then regularly trigger a new download of the datasets.