Warped Virtualizarr

[sharkinsspatial]

One of the prior art inspirations for Virtualizarr was the GDAL VRT concept which is analogous to several Virtualizarr virtual datasets which have been concatenated. GDAL also has the concept of Warped VRT. A Warped VRT contains not just the definition of the virtual references but also the information on how to apply the appropriate warp transforms between the constituent reference datasets and the CRS of the concatenated "parent" dataset.

We've had lots of discussion in the past about how we might be able to achieve this type of behavior in the Zarr world but it seems like we are finally at a point with the ecosystem where this is possible.

1. Virtualizarr now provides us functionality to create "virtual" datasets like GDAL VRT.
2. The amazing work at the Zarr Summit 👏 on the zarr-conventions framework gives us a new, clear way to define flexible attribute schemas to store specialized information.
3. The proposed geo-proj convention provides a standardized way to define an array's CRS and transform.
4. Critically we can now also define a new convention on the "parent" concatenated array called something like geo-warp that informs the client to apply a warping operation from the source transform -> destination transform using something like pyresample anytime data is read from the "parent" concatenated array.

Obviously there is "a lot" of work to make this a reality but this would be hugely valuable to provide a seamless experience (pun intended) to downstream users.

[emmanuelmathot]

Love the idea. Could this unblock our Sentinel-1 issue @vincentsarago?