Add SpaceM datasets

[aeisenbarth]

This PR adds two SpaceM datasets as examples for metabolomics.

SpaceM datasets natively use the SpatialData format, so actually only downloading is needed but no conversion. Still, I kept the pattern of both download.py and to_zarr.py for consistency.

Work still needed:

• (BLOCKER) Datasets were generated with spatialdata 0.1.2 and use . in element names, which is incompatible with the current version. Naming constraints break compatibility with existing datasets scverse/spatialdata#624 needs to be solved first, or the elements in the dataset need to be renamed.
• Transfer datasets to persistent, public storage. (I shared the datasets with @LucaMarconato .)
• Update URLs in the both download.py scripts
• Test the scripts in the workflow

[LucaMarconato]

Thanks for the contribution. I have uploaded the data to S3, updated the scripts and tested them (I have allowed . locally to test the scripts). I have also tried running the notebook on the subset data, it works.

[LucaMarconato]

I'll merge the PR.

[LucaMarconato]

One comment, independent of the PR. I have noticed that the labels show some interpolation effect at the border. Using transform, using the APIs from multiscale-spatial-image or using the rasterization APIs this should not occur. In case you used them you may have discovered a bug. Or did you use external methods?

[aeisenbarth]

Oh! This is a mistake in the data that I will have to fix. It seems only the small datasets were affected. Generated them by creating a new SpatialData object with only the scale1 labels (skipping scale0), and letting SpatialData build the pyramids.

I will inspect whether there is a bug, or whether I should have taken a different approach.

[aeisenbarth]

@LucaMarconato, I reuploaded the fixed datasets to scratch. When you have, can you replace them? Thanks!

[LucaMarconato]

Great! Done 😊