LLF MHD zeroes passive scalar fluxes

## Summary `LLF_MHD(...)` computes passive-scalar fluxes from conserved containers whose scalar arrays were never populated from the input states. ## Severity `High` ## Affected File `src/hydro/LLF_mhd.hpp` ## Affected Function / Symbol `quokka::Riemann::LLF_MHD(...)` ## Audit Metadata - Source log: `issues/likely_real/llf-mhd-passive-scalar-flux-zero.md` - Finding tags: correctness/MHD/passive-scalars ## Proposed Patch - Copy passive-scalar state from `sL`/`sR` into `u_L`/`u_R` before computing LLF fluxes. ## Why This Is a Bug `quokka::Riemann::LLF_MHD()` builds `u_L` and `u_R` from the left/right MHD states, then computes passive scalar fluxes from `u_L.scalar[n]` and `u_R.scalar[n]`. Unlike `HLLD()`, this function never copies `sL.scalar[n]` or `sR.scalar[n]` into `u_L.scalar[n]` and `u_R.scalar[n]`. The `ConsHydro1D` objects are value-initialized, so their scalar arrays remain zero. As a result, every passive scalar component in the MHD LLF flux is zero. This is high severity because `LLF_MHD` is used by the first-order hydro flux path (`hydroFOFluxFunction`) for MHD. Whenever first-order flux correction or retry logic uses this fallback in an MHD problem with passive scalars, scalar advection can be silently removed at corrected faces. ## Complete Code Patch ```diff diff --git a/src/hydro/LLF_mhd.hpp b/src/hydro/LLF_mhd.hpp --- a/src/hydro/LLF_mhd.hpp +++ b/src/hydro/LLF_mhd.hpp @@ u_R.Eint = sR.Eint; u_R.by = sR.by; u_R.bz = sR.bz; + for (int n = 0; n < N_scalars; ++n) { + u_L.scalar[n] = sL.scalar[n]; + u_R.scalar[n] = sR.scalar[n]; + } //--- Step 2. Compute wave speeds in L,R states (see Toro eq. 10.43) ```