Bonsai Llamafiles bundle 1-bit models into executables

This is less a new model than a deployment win: it turns a fork-fragmented Bonsai stack into a single-file local runtime that people can actually use. The real value is distribution simplicity, not new model capability.

• –The repo bridges PrismML's custom `llama.cpp` quantization support into llamafile, removing the need to juggle incompatible forks.
• –CPU-only packaging is the right default for a local 1-bit model; it keeps the pitch focused on portability and low-footprint inference instead of GPU chasing.
• –The 1.2 GB 8B executable is small enough to feel like a true download-and-run artifact, which is where llamafile still has a clear niche.
• –This is especially relevant for enterprise laptops and offline workflows, where shipping one binary is easier than managing runtimes and model installs.
• –The repo is a practical companion to PrismML's official Bonsai releases, not a replacement for them.