predict peak shape in q space

At the moment we estimate peak shapes by averaging in detector space, which is clearly wrong because it does not take into account distortion from detector geometry. Instead, we should transform the ellipsoids into q-space, perform the averaging in q-space, and then transform back to detector space.