Full `anndataR` support for Zarr V2/V3

[Artur-man]

Once scverse/anndataR#190 completes, basilisk/reticulate support either should be removed or moved (to SpatialData.data ?)

[HelenaLC]

@Artur-man so my understanding is that we can remove basilisk/reticulate entirely once anndataR builds on the BBS, or am I missing sth?

[Artur-man]

I should test it because last time I tried to read the VisiumHD with full anndataR I have waited for minutes. The only non-lazy part of SpatialData is perhaps anndataR, we do not have a ZarrSparseMatrix yet.

[HelenaLC]

I think we identified why that is with Hugo. It's to do with Rarr, not anndataR, and he will look into it.

[Artur-man]

ok ... but lets test with SpatialData.data examples first, then we can remove.

[Bisaloo]

Huber-group-EMBL/Rarr#161 & Huber-group-EMBL/Rarr#162 should help quite a lot with some of the performance issues we were seeing.

It doesn't yet make everything instant. Using ZarrArray in anndataR would definitely help a lot.

Huber-group-EMBL/Rarr#166 should also help quite a lot for the specific case of non-lazy zarr stores used in anndataR but it still needs more work before it can get merged.

[Artur-man]

@HelenaLC did you test all datasets in SpatialData.data ?, it can be slow thats ok I think ... if so I will remove basilisk completely.

[HelenaLC]

The Visium HD ones I was not able to load. One is huuuge and I gave up after 20', the other (mouse brain) gave an error. All other data were fine as far as I recall.

[Artur-man]

Yeap, VisiumHD we cant, in my opinion it requires DelayedSparseMatrix, ok then I will remove basilisk, it has been too long anyways