Use compact offsets+indices neighbor-list representation and compressed index type

[AhmedSalih3d]

Motivation

• Replace per-cell dynamic Vector{Int} neighbor lists with a contiguous offsets+indices representation to reduce allocations and improve cache locality and iteration speed.
• Use a compressed integer index type for neighbor references and add a CellIndexMap for fast cell lookup in boundary/MDBC routines.

Description

• Introduce NeighborCellIndexType = UInt32 and helper CompressIndex for safe index compression and export the new type via SPHNeighborList imports.
• Add a new CSR-like builder BuildNeighborCellLists!(NeighborCellListOffsets, NeighborCellListIndices, FullStencil, UniqueCellsView, ParticleRanges, CellIndexMap) and an overload that fills a CellIndexMap for the legacy API, while retaining the original BuildNeighborCellLists! behavior where needed.
• Replace NeighborCellLists usages with NeighborCellListOffsets/NeighborCellListIndices and update neighbor iteration sites in SPHCellList (NeighborLoopPerParticle!, SimulationLoop, MDBC flow) to iterate by cursor over offsets and to Int-cast compressed indices when needed.
• Change MDBC neighbor lookup to use an inline GhostCellIndex function and CellIndexMap get lookups instead of FindCellIndex, and update ApplyMDBCBeforeHalf! to pass the new arguments; update ComputeCellNeighborCounts to support the new offsets+indices layout.
• Update allocation/initialization sites in FinalizeSimulationOutput! to create NeighborCellListOffsets, NeighborCellListIndices, and NeighborCellIndexMap and wire them through the simulation loop and neighbor-list rebuilds.

Testing

• Ran the package test suite and the repository's simulation smoke test; all automated tests completed successfully.

---

Codex Task