More physically motivated and faster kernel functions
Two improvements for the kernel functions:
- Unroll the loops in kernel.h and don't use arrays for coefficients for better vectorisation.
- Use the FWHM definition of Dehnen+2012 to maintain resolution between kernel choices.