For transit geometry, is there any advantage in having an impact parameter sampling different than the inverse atmospheric layer sampling?
If not, we should automate the code to always use the inverse radius layering. This will greatly simplify the code in many places (transit.c, makesample.c, tau function in tau.c, ...).