Bugfixes in HPC SN Solver

[ScSteffen]

PR: Fix QOI/output correctness and probing geometry in SNSolverHPC

Summary

This branch fixes multiple correctness bugs in src/solvers/snsolver_hpc.cpp affecting MPI reductions, RK2 consistency, probing geometry, restart-time continuity, and lattice perimeter tagging.

Why

master had several issues that could skew reported QoIs or produce inconsistent behavior across run modes:

• MPI guards used BUILD_MPI in places where the project defines IMPORT_MPI.
• MAX_OUTFLOW was not reduced globally across ranks.
• RMS_FLUX under MPI used rank-local partial fluxes, making values rank-count dependent.
• In RK2, _scalarFlux could be out of sync with the final Heun-averaged _sol used for postprocessing.
• VAR_ABSORPTION_GREEN screen output used the wrong field.
• Green block probing rectangles were hardcoded around origin and ignored configured green-center offsets.
• Probing-line generation could hit invalid point-count cases (linspace2D denominator and odd probe-count layout assumptions).
• Restarted runs reset SIM_TIME to zero instead of continuing from the loaded iteration.
• Lattice perimeter checks had malformed abs(bool) logic.

What changed

File changed: src/solvers/snsolver_hpc.cpp

1. MPI guard consistency

• Replaced #ifdef BUILD_MPI with #ifdef IMPORT_MPI in this file so MPI barriers/reductions compile and run under the project’s actual MPI macro.

2. Global max outflow across ranks

• Added MPI_Allreduce(..., MPI_MAX) for _curMaxOrdinateOutflow in IterPostprocessing().

3. Correct RMS flux computation under MPI

• FVMUpdate() snapshots previous global scalar flux.
• RMS now uses global updated scalar flux minus previous global scalar flux.

4. RK2 scalar-flux consistency fix

• After Heun averaging of _sol, _scalarFlux is recomputed (and allreduced in MPI) before postprocessing.
• Ensures QoIs are computed from the final RK2 state.

5. Fix wrong screen field for green absorption variance

• VAR_ABSORPTION_GREEN now outputs _varAbsorptionHohlraumGreen (instead of _absorptionValsBlocksGreen[0]).

6. Green probing geometry alignment

• Shifted green block probing rectangles by _centerGreen (cx, cy) so block QoIs track configured center offsets.

7. Probe-count safety guards

• Added assert( num_points > 1 ) in linspace2D.
• Added asserts in SetProbingCellsLineGreen() for even total line probe count and minimum side counts.
• Added #include <cassert>.

8. Restart simulation-time continuity

• Initialize _curSimTime from _idx_start_iter * _dT in Solve() so resumed runs continue timeline correctly.

9. Lattice perimeter logic fix

• Corrected malformed perimeter comparisons and tightened branch structure (else if) to avoid incorrect overlap/classification.

Behavioral impact

• QoIs are now consistent across MPI rank counts for MAX_OUTFLOW, RMS_FLUX, and MOMENTS reductions.
• RK2 postprocessing uses fluxes consistent with final averaged state.
• Green block/line probing follows configured center offsets and enforces safe sampling assumptions.
• Restarted runs report continuous simulation time.
• Lattice perimeter-derived outflow metrics better match intended geometry.

Risk / compatibility

• No public API changes.
• Changes are confined to solver internals and output/postprocessing behavior.
• Assertions are disabled in NDEBUG; release behavior remains non-terminating for those guards unless converted to runtime errors.

Validation

• Manual code-path validation via targeted review.
• No automated test run included in this branch.