Cache the dowloaded Zenodo data in GitHub actions

[tcompa]

The advantage here (as compared to caching poetry environments, which we also do) is that:

1. These are small datasets, so that their contribution to the GitHub cache size limit won't be relevant
2. Their key is simply a DOI, and they are common to all tests (pip/poetry, core library vs tasks, different OSs, different Python versions...).

To do:

• [ ] Prepare a command that downloads the Zenodo record in the appropriate form and into the appropriate relative path (in tests/data subfolders)
• [ ] Prepare a reusable GitHub action with on: [workflow call], which simply downloads the two datasets (images and zarrs)
• [ ] Add the caching layer to this action
• [ ] Include it as a step in all other CI actions

Note that something similar should be considered also in view of #565.