Optimize neighbor list scratch usage

[AhmedSalih3d]

Motivation

• Reduce per-iteration allocations and redundant checks when building neighbor lists and producing grid cell statistics by reusing scratch buffers and precomputing cell occupancy.

Description

• Add NeighborListScratch (fields CellCounts, CellOccupied, NeighborCounts) and ResizeNeighborListScratch! to hold reusable cell/neighbor buffers in src/SPHNeighborList.jl.
• Add in-place helpers UpdateCellCounts!, ComputeCellParticleCounts!, and ComputeCellNeighborCounts!, and adapt ComputeCellParticleCounts/ComputeCellNeighborCounts to call the in-place variants in src/SPHNeighborList.jl.
• Change BuildNeighborCellLists! to accept a CellOccupied array and use occupancy checks (and sizehint!) instead of reading particle ranges for empty-cell checks in src/SPHNeighborList.jl.
• Wire a NeighborScratch instance into the simulation: create NeighborScratch = NeighborListScratch() in RunSimulation, call UpdateCellCounts! before BuildNeighborCellLists!, and pass NeighborScratch through SimulationLoop to reuse buffers for constructing neighbor lists and preparing VTK grid outputs in src/SPHCellList.jl.
• Re-export the new types and functions from SPHNeighborList and update the top-level SPHExample.jl exports accordingly.

Testing

• No automated tests were run on these changes.

---

Codex Task