Fix NEMO history MAX_/BGS_ residual groups: species-first indexing and finite maximum residuals

[bellonarts]

Proposed Changes

NEMO's MAX_RES and BGS_RES groups used the calorically perfect variable layout even though the solution vector starts with one density per species. For AIR-5, fields labeled as momentum and energy therefore read species residuals, and the actual momentum, energy, and vibrational-electronic residuals were missing. This change registers and loads the complete species-first layout in 2D and 3D. RMS/MAX/BGS share one loading lambda, with energy at nSpecies + nDim. Species loop counters are local and use auto iSpecies = 0u.

CSolver::SetResidual_RMS floors maximum residuals at EPS*EPS after the value and location reductions, matching the RMS convention and preventing log10(0) in output. This shared reducer also applies to non-NEMO solvers. BGS history uses the already-floored Residual_BGS; the distinct Residual_Max_BGS storage remains unchanged.

Breaking NEMO field-name change: replace MAX_DENSITY and BGS_DENSITY with explicit MAX_DENSITY_i and BGS_DENSITY_i selectors in CONV_FIELD, HISTORY_OUTPUT, and SCREEN_OUTPUT. Update relative and zone-qualified selectors similarly, for example REL_MAX_DENSITY_0 or BGS_DENSITY_0[0]. The old unsuffixed fields sampled species zero, not total mixture density. List every species that must satisfy the stopping criterion. New vibrational-electronic fields are MAX_ENERGY_VE and BGS_ENERGY_VE.

Obsolete NEMO CONV_FIELD names now produce an actionable startup error instead of being discarded and potentially weakening a mixed stopping condition. The check requires a registered per-species replacement, preserving valid non-NEMO names. Screen/history requests retain the existing unavailable-field handling.

The residual assembly and solution-update formulas are unchanged. Corrected group membership changes AVG_MAX_RES and AVG_BGS_RES, including zone-qualified averages used by multizone convergence monitoring. AIR-5 groups now include nine variables in 2D or ten in 3D, so runs monitoring corrected fields or averages can stop at different iterations.

Validation

Current head: d193edebce. This follow-up only changes the 13 species-loop declarations to unsigned auto, as requested in review. Recompiled current production/test objects passed 547 assertions in 4 focused cases on macOS arm64, non-MPI. This was an incremental diagnostic using existing native libraries.

The complete RunPod matrix below was run at parent 24162ed571, before that loop-style follow-up. A clean Linux x86_64 build used GCC 13.3.0, MPI and MLPCpp enabled. All 9,458 packaged source/dependency files matched their recorded hashes; retained build/test artifacts were retrieved and SHA-256 verified.

• Focused output/convergence suite: 547 assertions in 4 cases per rank, passing as a singleton and with 2 and 4 MPI ranks. Covers 2D/3D RMS/MAX/BGS indexing, single-zone and multizone loading, zero maxima, rank-dependent reductions, and maximum point/location checks.
• Explicit first/last species and energy criteria survive preprocessing and independently control convergence. Relative and zone-qualified names are covered. Unzoned relative checks evaluate the predicate after fields refresh; they do not change the existing monitoring/postprocessing order.
• All 9 migration probes passed as singletons and with 2 MPI ranks; the Meson migration test passed. Obsolete base, relative, zone-qualified, and mixed selections fail with explicit-species guidance. Normal-unit CI now invokes this runner explicitly.
• Full normal unit suite: 74,913 assertions in 44 cases passed.
• Official two-rank regressions passed at unchanged 1e-5 absolute tolerance: NEMO inviscid wedge/AUSM at iteration 10 and Euler channel/RK at iteration 20. The wedge emitted a warning about 20,768 nonphysical reconstructed upwind states. These are bounded reference-regression passes, not convergence or physical-admissibility claims.
• The clean build emitted zero compiler warnings. Direct checks with clang-format 15.0.7 and Black 22.6.0 passed.

OpenMP, AD, DD and sanitizer builds were not tested on RunPod. Fresh upstream checks at the current head remain the merge gate; earlier-head green checks are not treated as validation of this head. Original cold-start AIR-5 and QuickStart comparisons remain historical evidence and were not rerun for this revision.

Related Work

This PR repairs NEMO residual output indexing and maximum-residual reporting, and protects convergence configuration during the field-name migration. It is independent of the mass-fraction tolerance change in #2880 and interrupt-exit reporting in #2881.

PR Checklist

• I am submitting my contribution to the develop branch.
• The clean Linux build emitted no compiler warnings.
• My contribution is commented and consistent with SU2 style.
• I used pre-commit run --all. Direct pinned formatter checks were used for this revision; the complete hook was not run locally.
• I have added test cases for this contribution.
• The renamed fields, configuration migration, and stopping-criterion impact are documented above.

[bellonarts]

I updated this with the latest develop and all the checks are passing. The tests cover the MAX and BGS mapping for each species and confirm finite output when a residual is zero. I also clarified which history fields were renamed and which configs need updating.

[bigfooted]

If you update again I go through it and we can probably merge this week.

[bellonarts]

Updated with the latest develop, and all checks are now passing. Thanks for rerunning the affected jobs.

[bigfooted]

LGTM
@joshkellyjak can you do a final check before merging?

[joshkellyjak]

Just the SetHistoryOutputFields helper function change and then LGTM