Separate xgrid and model generation within model trainer

Created by: scarlehoff

Right now the xgrid that is used as input for the PDF (which is a concatenation of all the points in x the fktables ask for) was being prepared just before the observable generation.

There was no reason for that, it just happened to be there. For #1622 (closed) we need to generate the photon in all those x-points. While technically it doesn't matter where luxqed is called, I think it is better to call it from the top(est?) possible level. In this case, just after the grid of x-points is prepared.