rebuild

stop tracking build/, like the other ggml kernels

rebuild metal for get_rows, and track artifacts by glob

trim the metal shader to the dispatched kernels

Rebuild CUDA variants with the current builder (provenance, digest, no shim)

Refresh the Metal builds for the 20-type table

Refresh the Metal builds after the command-buffer fix

Add prebuilt Metal variants for torch 2.12 and 2.13

GGUF kernels: dequantize + fused gemv over packed blocks

GGUF kernels: dequantize + fused gemv over packed blocks, 9 CUDA variants

initial commit