Small bug in grid_noro fixed (smoothed topography was used instead of unsmoothed one for geopotential computation at north pole).
Removed average of mass at poles in etat0dyn_netcdf after start_init_dyn => different results in the zoomed grid case.
ok_etat0=n and ok_limit=y combination now works fine (if no initial state is needed, but only limit.nc file). This required:
to move grid_noro0 and start_init_noro0 subroutines from etat0dyn_netcdf.F90 to limit_netcdf.F90
to create init_ssrf_m.F90 file, so that sub-surfaces can be initialized from limit_netcdf.F90 without any etat0*_netcdf routines call).
Simplified somehow the corresponding code, in particular:
1) removed obsolete flags "oldice".
2) removed flag "ibar": barycentric interpolation is used everywhere (except in start_init_subsurf, still calling grille_m - to be changed soon).
3) removed useless CPP_PHY precompilation directives, considering the possibility to run ce0l without physics is useless (ce0l is dedicated to Earth physics).