3D Jacobi Stencil on Blackhole (CARLA 2026)

Reference implementation for the CARLA 2026 paper "Shift Matrices and Single-Batch Halos: Optimizing 3D Stencils on Tenstorrent Blackhole Architecture". A TT-Metalium 3D Jacobi stencil kernel for the p150a that uses shift matrices and single-batch halo exchange, with build and run scripts for reproducing the paper's results.