Parakon runtime: llama.cpp fork with fused 1-bit CUDA/Metal/CPU kernels (part 2)

Parakon runtime: llama.cpp fork with fused 1-bit CUDA/Metal/CPU kernels