…oblem domain

It may make sense to update the problem domain if possible

init_terrain_grid() used to use ztop = geom.ProbHiArray()[2], which
comes from the input file. For a stretched grid or user-specified
"terrain_z_levels", there could have been a mismatch between ztop and
the actual requested top of the domain.

…defintions in most prob.cpps

Note: The regression tests DynamicRefinement, IsentropicVortex, and
ScalarAdvDiff all explicitly initialized z_phys_nd to be k*dz and then
called FillBoundary(). However, if use_terrain==true (and
init_custom_terrain() was called), then the default behavior should be
equivalent: init_zlevels() called without grid stretching, then
init_terrain_grid() called to populate z_phys_nd and then FillBoundary()

…grid_stretching_updates

…eanup

Notes:
- Nominal z levels (e.g., from input terrain_z_levels) that are used for
  terrain smoothing should range from 0 (on the surface) to ztop
- The same levels are used to calculate the base state by integrating
  the hydrostatic equation through a column of air
- In the case of terrain or grid stretching, the base state profiles
  may have non-uniform vertical grid spacing
- These 1-D profiles are then interpolated to the 3-D terrain grid; the
  state and hse fab arrays are interpolated at z_phys_cc whereas the
  velocity fab arrays are interpolated at face centers calculated from
  z_phys_nd