Use cell-index map for neighbor lookups and refactor MDBC neighbor loop

[AhmedSalih3d]

Motivation

• Replace expensive search-based neighbor lookups with a direct map to speed up neighbor cell discovery and reduce allocations.
• Simplify and harden the MDBC ghost-point neighbor accumulation by avoiding repeated searches and early-skipping empty cells.
• Improve reuse and sizing of neighbor containers to reduce allocation churn during neighbor-list rebuilds.

Description

• Add a CellIndexMap / NeighborCellIndexMap (Dict{CartesianIndex,Int}) to map cell coordinates to their index and thread it through BuildNeighborCellLists!, NeighborLoopMDBC!, ApplyMDBCBeforeHalf!, and SimulationLoop.
• Overload BuildNeighborCellLists! to populate and size-hint the CellIndexMap and NeighborCellLists, and switch neighbor lookups to get(CellIndexMap, cell, 0) to avoid searchsortedfirst in hot loops.
• Modify NeighborLoopMDBC! to use an inlined GhostCellIndex helper instead of the previous f, iterate FullStencil, and skip missing neighbors early; return per-ghost contributions using ComputeInteractionsMDBC! as before.
• Update ApplyMDBCBeforeHalf!, SimulationLoop, and the top-level run logic to create and pass NeighborCellIndexMap and to call the new BuildNeighborCellLists! signatures; remove unused GhostNormals unpack and other small cleanups.

Testing

• Ran the package unit test suite with ] test against the modified code and observed all tests pass.
• Executed a short smoke simulation run (example input) to validate neighbor-list rebuilds and MDBC corrections, and it completed without errors.

---

Codex Task