Add a high-level method to assemble single-cell data with AnnData

[arteymix]

TODO:

• concatenate multiple datasets
• create a multi-index (dataset_id, sample_id, cell_id), the dataset-level can be omitted if a single dataset is requested
•  include factors, sample characteristics and cell-level characteristics
• deal with multiple platforms (I think we could use the design element ID for that, or a multi-level index)
• generalize the concept of download_dir and implement some form of invalidation, I think that the Last-Modified is exposed by the backend and there might even be Cache-Control headers we could leverage there.

Depends on #38 because I needed authentication for the test dataset. The final PR will use a public dataset.

[arteymix]

Looks like the numerical ID issue is being addressed in scverse/scanpy#3909. We'll have to set the minimum version of scanpy accordingly.