Complete named-constant adoption and parameter declaration generation

[sbryngelson]

Summary

Follow-up to #1550, completing two more pieces of the same strategy — one source of truth in the parameter registry, generated everywhere, no behavior change:

1. Named-constant adoption finishes. The 9 remaining enumerated vocabularies (bubble_model, avg_state, wave_speeds, recon_type, muscl_order, muscl_lim, int_comp, format, precision) are swept across all three executables — 99 decision sites move from integer literals to the generated named constants. The legacy WENO_TYPE/MUSCL_TYPE constants are retired in favor of the generated recon_type_weno/recon_type_muscl (same values, one naming scheme).
2. The params pipeline now generates ALL namelist-bound declarations. A TYPED_DECLS table in definitions.py drives generation of the 14 derived-type namelist declarations (fluid_pp, patch_icpp, acoustic, …) per target — including their GPU_DECLARE lines and Doxygen descriptions — and the simulation case-optimization #:if MFC_CASE_OPTIMIZATION block is emitted from a registry loop instead of being hand-maintained. The corresponding manual declarations in the three m_global_parameters.fpp are deleted (net −40 LOC of hand-synced Fortran). Adding a parameter no longer requires touching Fortran declarations at all; adding a case-optimization parameter no longer requires editing a parallel #:else block.

Two infrastructure hardenings that fell out of the work:

• Generated includes regenerate at build time. Previously they were written only at CMake configure time, so a cached build that skipped reconfigure compiled stale includes (a silent wrong-binary hazard we hit during development). A ninja-tracked add_custom_command now owns regeneration: touch definitions.py && ninja does the right thing without reconfigure.
• Doc lint accepts registry-derived tokens. lint_docs.py validates backticked tokens in case.md against the registry's value names and the analytic-expression intrinsics instead of a hand-grown skip list.

Verification

• Every Fortran-touching commit was gated on the full golden suite (sharded across two parallel jobs) with zero golden-file regenerations, four full-suite runs in total including one against current master after Unify ICPP STL onto the shared IB model path #1546 landed.
• The declaration-generation steps were additionally proven by declaration-set equivalence on the Fypp-emitted .f90: every declaration and GPU-residency directive before the deletion is byte-identical (modulo relocation and comments) to the generated form after, per target and — for the case-optimization block — in both MFC_CASE_OPTIMIZATION configurations.
• Toolchain pytest suite extended to 323 tests; precheck green on every commit.

What still requires manual edits (documented in .claude/rules/common-pitfalls.md)

Derived-type member definitions in m_derived_types.fpp, default-value assignments, the six case.py-computed case-opt extras, multi-variable declaration lines (bc_x/y/z, domains), and the MPI broadcast lists — the broadcast lists are the next generation target.

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

[codecov]

Codecov Report

❌ Patch coverage is 23.95833% with 73 lines in your changes missing coverage. Please review.
✅ Project coverage is 61.17%. Comparing base (c0792ec) to head (12dfa2a).
⚠️ Report is 1 commits behind head on master.

Files with missing lines	Patch %	Lines
src/post_process/m_data_output.fpp	5.00%	7 Missing and 12 partials ⚠️
src/simulation/m_riemann_solvers.fpp	25.00%	4 Missing and 5 partials ⚠️
src/common/m_mpi_common.fpp	0.00%	0 Missing and 7 partials ⚠️
src/simulation/m_qbmm.fpp	50.00%	0 Missing and 5 partials ⚠️
src/simulation/m_bubbles_EL.fpp	0.00%	0 Missing and 4 partials ⚠️
src/simulation/m_muscl.fpp	0.00%	0 Missing and 4 partials ⚠️
src/post_process/m_mpi_proxy.fpp	0.00%	1 Missing and 2 partials ⚠️
src/pre_process/m_data_output.fpp	0.00%	0 Missing and 3 partials ⚠️
src/simulation/m_cbc.fpp	25.00%	0 Missing and 3 partials ⚠️
src/simulation/m_data_output.fpp	0.00%	0 Missing and 3 partials ⚠️
... and 8 more

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #1551   +/-   ##
=======================================
  Coverage   61.17%   61.17%           
=======================================
  Files          74       74           
  Lines       20313    20313           
  Branches     2961     2961           
=======================================
  Hits        12427    12427           
  Misses       5870     5870           
  Partials     2016     2016           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[sbryngelson]

Code review

No issues found in this PR's own commits. Checked for bugs and CLAUDE.md compliance across five independent passes (compliance, shallow bug scan, git-history conflicts with recent master merges, prior-PR feedback, and in-code guidance), with emitted-declaration equivalence verified per target. The codecov patch-coverage number reads low because the diff is dominated by generated Fortran declarations, which the golden regression suite (not unit tests) exercises; the toolchain Python in this PR is unit-covered.

🤖 Generated with Claude Code

_{- If this code review was useful, please react with 👍. Otherwise, react with 👎.}