Canonical RANS CTests: run without FFT in CI; terrain-height wall distance on z-split layouts

[hgopalan]

Canonical RANS CTests: run without FFT in CI; terrain-height wall distance on z-split layouts

Hazard-side copy of the two fixes pushed to erf-model#3971 today (cherry-picks 3eff4c7 and 9a9207f, plus the RANS_Neutral_ABL_Flat_Implicit entry that only exists here).

Why

No CI configuration builds with ERF_ENABLE_FFT, and every Canonical_RANS deck ran the anelastic projection with the FFT solver, so all ten RANS entries merged in #405 abort at start-up in CI: the flat decks in ReadParameters on erf.use_fft, the 2D hill deck (which already sets use_fft = false) in project_momenta, because the general-terrain projection is the FFT-preconditioned GMRES with no non-FFT path. The harness also swallowed the run log, so ctest -VV showed nothing. The upstream PR's CI run of 2026-09-10 failed 9 of 9 on macOS, Linux and Windows; every later run was cancelled or queued, so it went unnoticed.

Changes

• The four flat entries pass erf.use_fft=false (MLMG projection) and run in every build; the six terrain-fitted entries are registered only under if(ERF_ENABLE_FFT).
• The test command echoes the tail of the run log when the executable exits non-zero.
• erf.wall_dist_type = terrain_height read the surface nodes z_nd(:,:,klo) from every box, but boxes above a z split of the BoxArray do not hold them (778 m error on the 3D hill at amr.max_grid_size_z = 10). The surface slab is now gathered onto every box first, with a guard that reduces over the slab's own ghost width.
• PLAN.md phase 6, the 3D hill README, RESULTS.md, the deck and the Poisson-solver comment no longer attribute the 1.788e139 pre-projection divergence to dz != dx; it was the z-split layout (Fix momenta ghost faces and planar surface-layer arrays on z-split BoxArrays (terrain-fitted meshes) erf-model/ERF#3970). The Poisson wall-distance multigrid does still diverge at dx = 2 dz.

Verification

• Debug build without FFT: 4 entries registered under -L rans, 4 pass (the three Timestep_Limits dt-sweep entries, whose spin-up deck also uses FFT, are now gated the same way).
• Release FFT build: 13 entries pass (10 regression plus the 3 dt-sweep).
• 3D hill, max_step = 0, amr.max_grid_size_z = 10 vs one box per column: wall distance bit-identical with the fix (778 m apart without it); the unsplit distance is unchanged.
• Codespell clean; the trailing-whitespace lines in CTestList.cmake predate this branch.

🤖 Generated with Claude Code