AGRIF update following tests in IMMERSE realistic configurations

1. Revised preprocessing tool (DOMAINcfg):

• With the same vertical grid: Build coarsened grids over parent area compliant with a volume weighted feedback of tracers. Adjust centers of mass (i.e. "gdept" or "e3w") to correctly feedback pressure at the same time. Take into account variable cell areas as found with curvilinear grids to match volume and cross-sections.
• Vertical remapping: accommodate different vertical z grids. Exactly match volume everywhere which can lead to small thicknesses (smaller than what namelist options would lead to - these are located in ghosts areas and should not lead to growing errors). Nesting s into z (not VQS into z) is formally possible but requires some more work for realistic applications (dealing with bathymetry smoothing at the child-parent interface in particular).
• Introduced "ln_dept_mid" namelist option (now the default with AGRIF) which define cell centers at the middle of T-cells (instead of assuming an analytical function). This is the only approach compliant with a finite volume vertical remapping approach.

2. Various changes have been made to accommodate NEMO-AGRIF to the expected input meshes and the generalization of the "volume weighted" feedback of tracers.

3. Revert to "conservative" 1st order interpolation of tracers, barotropic transport and sea-level at child boundaries. This replaces the 2nd order interpolation used so far to prevent from errors with partial bottom cells and to ensure volume (divergence) conservation near the boundaries. Some more work is needed to restore 2nd order schemes (incl. Balsara's divergence conserving interpolation scheme) near the topography.

4. The resulting nesting scheme conserves volume, is constancy preserving and monotone. It is neutral to an horizontally homogeneous linear stratification (this last property is always true without vertical nesting, but requires turning off monotone limiters to do so with different vertical grids).