Support downsampled Z (/T/C)

[keller-mark]

User story

Potentially related to #669. Related to scverse/spatialdata#954. Related to vitessce/vitessce#2177

When SpatialData creates a multi-scale OME-Zarr image for a volume, it currently downsamples in the Z dimension. This trips up Viv, which currently assumes that any "selection" besides X/Y directly maps to a Z/C/T index in the underlying array.

Preferred solution

Viv should transform the Z/T/C-index "selection" to the underlying array index. For multiscale images, this means Viv should first check whether the Z/T/C dimension size is the same for all resolutions. If the dimension size for these dimensions is reduced (compared to resolution level 0) in the 1+ resolution levels, then assume downsampling occurred, and transform the selection index accordingly.

Possible alternatives

[keller-mark]

It would be kind of odd to downsample in the C-dimension, but I could see a use case for Z/T at least.