[MAGIC] Transformation resolving using vertices instead of pixel offsets.

[YooSunYoung]

The position vectors are zipped assuming there is x_pixel_offset but MAGIC decided not to have any pixel offsets in the file.
So the assumption that x_pixel_offset will always be present is now wrong.

https://github.com/scipp/scippnexus/blob/3550be53b4cb1897e82082400651fecd4a87fc58/src/scippnexus/nxtransformations.py#L493-L520

I asked Iurii in person and MAGIC defines each pixel's 8 voxels positions respect to the sample directly, which will always stay the same in terms of relative position (x, y, z). The only thing that changes is the scattering angle.

Then we need to skip the pixel offset and difectly use the voxel positions...?
i.e. if there are no pixel positions in the file, we just use position of the vertices.
@MridulS Do you maybe have more context on this...?

[jl-wynen]

Can you give an example how the voxels are encoded in a MAGIC file?

[SimonHeybrock]

If I remember correctly, for DREAM it was decided to store the pixel offsets since it consumed less space. Original intention was to rely on NXoff_geometry. This should be decided by MAGIC together with ECDC.

ScippNexus can already load NXoff_geometry. As NXoff_geometry does not really define what kind of shapes it is describing, there is likely no generic way of computing positions from that — it requires bespoke instrument knowledge, so would likely belong into ess.magic (if it is indeed decided to go with NXoff_geometry).

[YooSunYoung]

Then I'll move this to ess project instead.