Normalize sinogram data

(cherry picked from commit f8c35519)

This fixes uses on the reconstruction.

If the proj=dark=flat every values are NaN. As result everything fail.

That's the case in the lab, when using a real camera.

As the flatfield correction is supposed to return a normalized count, we can assume if there is nothing we can put 0. And anyway it is not data which have to be used for scientific use case.

That a normalization already used in the tomovis tiling service.