Updates for terrain / grid stretching

[ewquon]

• Fix bug w/ higher-order advection + terrain (ed6ae8c)
• Clean up problem-specific terrain code
• Warn if mismatched prob_hi[2] and zlevels_stag[nz] -- the latter is the stretched grid levels that are used to populate z_phys_nd and may differ from the input geometry if "grid_stretching_ratio" or "terrain_z_levels" is specified
• Use zlevels_stag[nz] in init_terrain_grid() instead of ProbHiArray[2] -- only has an effect if surface elevation z_arr(ii,jj,k0) != 0
• Enable terrain for idealized case with input sounding

[ewquon]

Draft for now, need to verify that the new default init_custom_terrain() doesn't break any reg tests...

[ewquon]

Note: regtests fail for DensityCurrent_detJ2 and DensityCurrent_detJ2_nosub (both with terrain, default 3rd order) fail starting from ed6ae8c, i.e., after the bug fix.

[ewquon]

302d57e enables ideal input soundings to be used with terrain, resulting in comparable pressure and density base states with and without grid stretching. The commit message describes the implementation. Some additional comments:

• The nominal grid levels, either supplied by the user through terrain_z_levels -- or grid_stretching_ratio and initial_dz -- are the "coordinate surface heights" ($\zeta$) from Klemp 2011
• $\zeta$ is expected to range from 0 (at the surface )to $z_{top}$
• With BTF and Sullivan TF terrain grids, the physical height of the highest grid nodes is exactly equal to $z_{top}$ at k=geom[0].bigEnd(2)+1, (where $\zeta=z_{top}$).
• With STF terrain grids (smoothing following Klemp 2011), the grid should be deformed near the surface only (see Implementation of terrain grid with STF smoothing #1368). The current implementation assumes the domain top is far enough away from the surface to be outside of the deformed grid region, in which case, the physical height of the highest grid nodes is also equal to $z_{top}$.

[ewquon]

This PR looks bigger than it actually is. I consolidated all the duplicated terrain initialization code (that set the surface z_phys_nd = 0) and cleaned up most of the prob source files.

@asalmgren a heads up that the gold data were updated!