OME-Zarr - precomputed affine transform mismatch

In HMBA and mouse stereotaxic atlases, the affine matrix is now different between precomputed and OME-Zarr. For example, in the HMBA human atlas, the template and annotation OME-Zarr signals overlay nicely, but the precomputed annotations have a translation mismatch (example: https://allen.neuroglass.io/studies/06a10e24-b65c-7af2-8000-84077b4399db).

This looks like it comes from the fact that the precomputed format only takes in scale parameters from the nifti affine header. The question is, do we 1) specify a translation offset to align precomputed to ome-zarr or 2) just apply the translation in the viewer state?

My preference would be to add a voxel offset to the precomputed `info.json` but not sure if that would be a breaking change.