Agrif and s-coordinates

[jchanut]

There is a problem with AGRIF and s-coordinates caused by the arbitrary value chosen for vertical scale factors over land (1m). Baroclinic velocity obcs in AGRIF are indeed written as a weighted average over 3 lines(columns) across the boundary. Since fluxes and not velocities are exchanged between grids, one need a meaningful value of vertical scale factors at the fine grid ghost points (ie at the outermost masked points of the domain).
To fix that, one can duplicate needed scale factors in domzgr. Note that in z-coord, the problem (almost) does not arise, because background scale factors are set to the reference values.