Fix the IBSEB build break and legacy surface-layer input detection

[asalmgren]

Two fixes on top of the #3860 merge.

1. development did not compile at HEAD

#3860 changed ERF::m_SurfaceLayer from a single unique_ptr<SurfaceLayer> to one
entry per domain face. #3960 added ibseb_advance, which reads the surface layer for
the wall function beyond neutral. The two merged independently, git merged them
cleanly, and the result does not build — ERF_IBSEB.cpp:143-146 still dereferenced
m_SurfaceLayer as a pointer.

Indexed to zlo: what the wall function needs is the ground beneath the buildings, and
that is the face the other zlo-only consumers already select (ERF.cpp:876).

2. Legacy surface-layer input detection was reading the wrong thing

With more than one surface-layer face, ERF::InitData_post decides whether to keep
reading the historical unqualified erf.most.* / erf.surface_layer.* keys for zlo by
asking whether any such key is present. It asked the live ParmParse table, which cannot
answer that question: ERF_InputSoundingData.H:150,165 queries most.surf_temp and
most.surf_moist with queryAdd while reading the sounding, and queryAdd inserts
the default it was handed when the key is absent. By the time InitData_post runs, both
keys sit in the table holding their negative sentinels whether or not the user typed them.

So every init_type = input_sounding run with more than one surface-layer face looked as
though it carried legacy inputs:

• a deck that also used the face-qualified spelling hit the "both spellings present"
  abort and could not start — including Tests/test_files/ABL_MOST_Cloudchamber, the
  regression case added by Extend SurfaceLayer for all boundary sides #3860;
• a deck without it silently took the legacy path.

The fix snapshots the inputs table at the top of ReadParameters() — the last point at
which presence in the table still means the user wrote it — and the detector scans the
snapshot. This is prefix-agnostic, so it stays correct if another erf.most.* queryAdd
appears elsewhere later, which a fix targeting just the two known keys would not.

Verification

ABL_MOST_Cloudchamber is its own regression test. Run locally with the anelastic solve
skipped (this build has no FFT support, so the unmodified deck dies at
project_initial_velocity before reaching the surface-layer code):

Case	Result
Cloudchamber deck, pre-fix binary	aborts: "Both legacy unqualified and zlo-qualified surface-layer inputs are present"
Same deck, fixed binary	initializes cleanly, no warning, no abort
Legacy unqualified erf.most.* + a second face	warning, run continues
Both spellings present	aborts with the intended message

erf_unit_tests: 585 passed, 3 skipped, 0 failed.

Docs

Docs/sphinx_doc/SurfaceLayer.rst now states the prefixing rule — unqualified for a lone
zlo face, face-qualified once a second face is declared, unqualified applied to zlo with a
warning, both spellings an error — and lists what actually aborts away from zlo: the
bulk_coeff / custom / rico flux types, prescribed surface heat flux, the adiabatic
case, variable sea roughness, and most.pblh_calc / most.include_wstar. The page
previously covered this only as "not all existing options are supported".

Note, not fixed here

In the legacy-warning case, erf.most.surf_temp / surf_moist are also read by the
input-sounding consistency check, which compares with exact floating-point equality — a
sounding line of 12.13 g/kg will not match most.surf_moist = 0.01213. Pre-existing and
orthogonal, but it makes the legacy path harder to exercise than it looks.

🤖 Generated with Claude Code