Localize support buffers in simulation loop

[AhmedSalih3d]

Motivation

• Reduce reliance on globally allocated support arrays by allocating per-run buffers inside the main timestep loop to better match the per-particle execution model.
• Avoid preallocating support structures in RunSimulation so memory lifetime and initialization are colocated with the simulation loop.

Description

• Allocate dρdtI, Velocityₙ⁺, Positionₙ⁺, ρₙ⁺, and shifting arrays inside SimulationLoop using @alloc and conditionally allocate shifting buffers when SMode is not NoShifting in src/SPHCellList.jl.
• Remove the call to AllocateSupportDataStructures from RunSimulation and simplify the SimulationLoop parameter list so those buffers are no longer passed in.
• Initialize the new local buffers (fill! for scalars and copyto! to prime Positionₙ⁺, Velocityₙ⁺, and ρₙ⁺) so loop logic continues to use the same temporaries as before.

Testing

• I ran repository inspection and edit commands used during the change: ls, cat AGENTS.md, rg --files -g 'AGENTS.md', sed -n '1,260p' src/SPHCellList.jl, sed -n '260,620p' src/SPHCellList.jl, rg -n "RunSimulation|SimulationLoop|HalfTimeStep|FullTimeStep|AllocateSupportDataStructures" src/SPHCellList.jl src/PreProcess.jl src/TimeStepping.jl, sed -n '640,940p' src/SPHCellList.jl, sed -n '1,260p' src/PreProcess.jl, sed -n '1,220p' src/TimeStepping.jl, rg -n "DensityEpsi" -n src, sed -n '1,120p' src/SimulationEquations.jl, rg -n "AllocateSupportDataStructures" -n src, rg -n "SimulationLoop\\(" src/SPHCellList.jl, git status --short, and git diff, and committed the change.
• No automated test suite or runtime tests were executed for this change.

---

Codex Task