1. Fix the mismatch between the psnbnd in coreprofiles and gray_core.
This happens whenever gray overrides the externally provided one
(i.e. the density tail would become negative before psnbnd and is so
rescaled to end exactly on the zero).
2. Make psnbnd no longer required by always computing it as in 1.
It hasn't been removed, because gray_params.data is sacrosant,
but it no longer has any effect.
3. Cleanup: mark public functions, restructure the global variables into
three categories; add comments explaining the analytical profiles
format, formulae and how the polynomial tail is computed.
