Ah, I see what you mean. So this is a bit tricky; the ordering of cells inside a data object (like a sphere, or a cutout) is not well-defined, so we return them as flattened arrays. In particular, when you have any different resolution points collected within the cutout/sphere, it's not obvious how to return these in 3D.
Is it sufficient to be able to get the x,y,z coordinates, in addition? Or, would it work to mask out the cells (with either 0 or NaN) in a 3D array that match up with the cutouts?