Add cell-pair (pairwise) neighbor loop and pairwise interaction helpers

[AhmedSalih3d]

Motivation

• Reduce redundant work by switching to a cell-pair / pairwise accumulation path so each i,j interaction is evaluated once and both particles are updated, giving ~2× savings for symmetric interactions in the single-threaded case.

Description

• Add NeighborLoopPairwise! variants that perform symmetric pairwise accumulation for the combinations of shifting and kernel-output modes, and zero/fill shared arrays before accumulation.
• Add ComputeInteractionsPairwise! and ComputeInteractionsPairwiseNoKernel! helpers that compute and return symmetric contributions for density diffusion, pressure/viscous acceleration, kernel output, and shifting terms in one call per pair.
• Update SimulationLoop to dispatch to the pairwise path when Threads.nthreads() == 1 and keep the original per-particle threaded path for multi-threaded runs.
• Use symmetric accumulation (add contributions to both i and j) and compute AccelerationMax from the final accumulated accelerations.

Testing

• No automated tests were run against the modified code.
• Commands executed while implementing and inspecting the change (all completed successfully): rg -n "ComputeInteractionsPerParticle" -S src, sed -n '320,620p' src/SPHCellList.jl, rg -n "compute_viscosity" -S src, sed -n '1,200p' src/SPHViscosityModels.jl, rg -n "compute_density_diffusion" -S src, sed -n '1,220p' src/SPHDensityDiffusionModels.jl, rg -n "Scratch|scratch|Thread" src/SPHCellList.jl src -S, sed -n '1,120p' src/PreProcess.jl, nl -ba src/SPHCellList.jl | sed -n '30,280p', nl -ba src/SPHCellList.jl | sed -n '660,980p', nl -ba src/SPHCellList.jl | sed -n '1260,1420p', git add src/SPHCellList.jl, and git commit -m "Add cell-pair neighbor loop path".

---

Codex Task