New Feature: Adding input file verification to OpenFAST, FAST.Farm and drivers.

[mayankchetan]

Feature or improvement description

New -CheckInput command-line option that verifies an input deck without running a simulation:

openfast -CheckInput MyTurbine.fst

The check runs the full initialization path of every module enabled by the input file — reading and
validating all module input files, resolving file references, building meshes and airfoil tables,
loading controller/interface libraries, and running the glue code's cross-module consistency checks
and mesh-mapping setup — then exits before any time marching. Three outputs:

• a self-contained console INPUT CHECK SUMMARY (per-module status, every error and warning with
  its source);
• a machine-readable <RootName>.verify.yaml report, written incrementally (a file without the
  trailing overall_status: block indicates the check itself crashed — a deliberate liveness
  contract for scripted callers);
• process exit code 0 (deck valid; warnings allowed) or 1 (one or more fatal input errors).

Unlike a normal run, which stops at the first fatal error, -CheckInput uses
attempt-everything error accumulation in the openfast glue: when one module's init fails, the
error is recorded and every remaining module is still attempted (with guarded/stubbed interface data
where a failed upstream module would otherwise cause invalid dereferences, disclosed in the report as
status unavailable). A deck with several independent problems is fully diagnosed in one
invocation — the motivating use case is scripted and agent-driven workflows that generate decks
programmatically, where today each error costs a full simulation launch. It also replaces the common
TMax = DT "one-timestep" validation hack.

The flag is available on every user-facing executable:

Executables	Check semantics
openfast	Full attempt-everything accumulation across all modules + glue cross-module checks
FAST.Farm	Per-component accumulation; the per-turbine init loop runs all turbines with T<n>: attribution; downstream farm steps after a failure reported as skipped
turbsim, 12 module drivers (aerodyn, aeroacoustics, hydrodyn, seastate, moordyn, inflowwind, aerodisk, sed, soildyn, orca, beamdyn, unsteadyaero)	Init-only check with report (fail-fast semantics; report suffix .driver.verify.yaml)
Excluded (documented in the user docs)	servodyn_driver (hardcoded test harness; deck is a code literal), feam_driver (no CLI), openfast_registry, C++ glue (dryRun already exists in the C++ API)

Prior art: the C++ API's dryRun flag; LS-DYNA mcheck=1; OpenFOAM checkMesh.

Related issue, if one exists

No existing issue or PR requests an input-check mode for the Fortran CLI.

Impacted areas of the software

• NWTC Library: new module NWTC_CheckInput.f90 (error collector, console summary writer,
  incremental YAML report writer, exit-code helper, driver-side helpers) + unit tests
  (fortran-lang test-drive, 13 tests).
• VersionInfo / CheckArgs: new CHECKINPUT flag case + unit test.
• openfast glue (FAST_Prog.f90, FAST_Subs.f90, FAST_Registry.txt/generated
  FAST_Types.f90): dispatch branch, new FAST_CheckInput_T driver routine, a CheckInputMode
  parameter, collect-and-continue behavior in FAST_InitializeAll's contained Failed() (gated on
  an optional collector argument), dereference guards/stub meshes for failed-module continuation,
  per-module report labels.
• FAST.Farm glue (FAST_Farm.f90, FAST_Farm_Subs.f90): dispatch + Farm_CheckInput,
  collector threading through Farm_Initialize. Also fixes three latent IsInitialized-set-on-
  failed-init bugs (AWAE, per-turbine WD and FWrap instances) that would have had FARM_End tear
  down failed instances.
• TurbSim (TurbSim.f90) and the 12 driver programs listed above (main files + their argument
  parsers; InflowWind_Driver_Registry.txt regenerated for one new flag field).
• Regression test infrastructure: new baseline-free runner reg_tests/executeCheckInputTest.py
  (fixtures generated at test time by copying + corrupting pristine r-test cases — the r-test
  submodule is not modified) and 19 CTest cases under the checkinput label.
• Documentation: docs/source/working.rst (usage, report contract, availability table,
  exclusions, DLL caveat, non-goals).

No changes to any module's physics or Validate* routines. All error detection reuses the
existing validators; the feature only changes what happens after they report.

Backward compatibility: without the flag, behavior is unchanged. All new code paths are gated on
the CheckInputMode parameter (default false) or the presence of the optional collector argument
(never passed by existing callers, including the C bindings / Simulink / AeroMap paths). The few
ungated allocated()/%committed guards are unreachable-when-false on normal runs (failure aborts
before those lines). Verified empirically: baseline regression tests pass (below) and byte-identical
output spot-checks were done per driver during development.

Additional supporting information

• The attempt-everything mode doubles as a failure-path exerciser for init code. A sweep of the full
  r-test corpus (173 cases) through -CheckInput surfaced two latent state-consistency flaws in
  existing code that fail-fast behavior had been masking; both are guarded in this PR and locked in
  as regression tests:
  • the glue read SrvD%Input(...)%ExternalBlPitchCom unconditionally after ServoDyn init (never
    allocated if SrvD_Init fails early);
  • SlD_Init's REDWIN setup sets a fatal ErrStat without an immediate return, so a subsequent
    successful allocation's error check trips on the stale fatal, leaving WriteOutputHdr allocated
    but WriteOutputUnt not; FAST_InitOutput trusted the header array alone. (The missing
    early-return in SoilDyn.f90 itself is a candidate for a small separate fix; this PR only
    hardens the glue, per its no-module-changes rule.)
• The same sweep found two r-test driver inputs that fail to parse as-shipped
  (ad_BAR_OLAF/ad_driver_Advanced.dvr: missing numCases; ad_Kite_OLAF/ad_driver_8figure.dvr:
  missing analysisType). Not addressed here — can open r-test issues if useful.
• Known limitations (documented): runtime-only checks (large-deflection warnings, solver
  convergence, NaN growth) are out of scope by design; TurbSim's pre-existing output-writability
  probe leaves a 0-byte .wnd placeholder during a check; MoorDyn's module-level .MD.out header
  is still written (module scope); modules whose init loads an external DLL (SoilDyn REDWIN,
  OrcaFlex) report a missing DLL as that component's failure — deliberate, since the DLL is part of
  the deck/environment contract.

Generative AI usage

Substantial portions of this feature were developed with Anthropic Claude (Claude Code), under human
direction for all design decisions. Specifically, a multi-agent workflow was used: Claude Fable 5
orchestrated the work (design/spec interview with the human author, implementation planning, and
adjudication of review findings); Claude Sonnet agents performed the code implementation, the
per-task independent code reviews, and the codebase reconnaissance; Claude Opus performed the two
whole-branch final reviews; Claude Haiku handled small fully-specified tasks (documentation,
single-file test fixes). Every task was gated by an independent review pass (spec compliance + code
quality) before proceeding, and all changes are backed by the test evidence below.

Co-authored-by: Anthropic Claude claude@anthropic.com

This PR heavily relies on OpenFAST Agentic Kit (OAK) workflows.

Test results, if applicable

All on macOS (arm64), gfortran 15.2, RelWithDebInfo, double precision:

• Unit tests: nwtc_library_utest 13/13 (new NWTC_CheckInput suite); versioninfo_utest 15/15
  (including the new flag case; no regressions).

• New feature tests: ctest -R checkinput — 19/19. Includes the multi-error acceptance test (two
  independent module errors seeded, both must be reported in one run), failed-module cascade and
  segfault-regression cases, FAST.Farm per-turbine attribution, and per-executable rollout cases.
  These tests are baseline-free (exit code + report assertions only) and generate corrupted fixtures
  at test time.

• Full r-test corpus sweep through -CheckInput (ad hoc, not CI): 173/173 cases complete without
  crashing — 168 pass, 3 fail cleanly with correct reports (1 macOS AMReX build limitation, the 2
  malformed r-test .dvr inputs noted above), 0 crashes.

• Baseline physics regression (unchanged-behavior check): AWT_YFix_WSt, AOC_WSt,
  5MW_Land_DLL_WTurb — all pass channel-by-channel against stored baselines on the feature
  branch (reference DISCON controllers built via regression_test_controllers).

• r-test branch merging required — not required: no files in the r-test submodule are added
  or modified; all new test fixtures are generated at test time from pristine copies.

• please squash PR before merging