Make inelastic sampling framework more flexible

[tkittel]

In order to work more easily on e.g. #40, UCN modes, improved Gn-based sampling at lowE, etc., it would be good if we generalised the concept of s scattering kernel a bit. Sometimes we really only have a tabulated pre-processed kernel, but sometimes we also have the ability to evaluate it without any sort of tabulation (i.e. we have Gn functions). Or perhaps we want to split it up (VDOS->SAB as always but leaving out G1..G3 and separate "kernels" for G1, G2, and G3). Or split up based on energies etc.

[tkittel]

If we expose the kernel evaluation capability to python, this would also solve #85.