this removes more code from the inner loop. It is more efficient,
but is it also more readable? Probably not.

the curvature correction makes that the edges of our integration
range are not straight. We need to integrate over a larger range
then the resulting spectrum would suggest. Before this code we tried
that, but actually failed.

this documents the hRIXS class, and does some minor code cleanup.

Move load_bpt and get_unique_sase_pId to bunchpattern.py, add npulses_has_changed() and get_sase_pId() functions

that was a quick hack that should not be done, background subtraction is
the way to go.