Enable containers on macOS to use the GPU

[slp]

Three changes:

• Bump llama.cpp to latest upstream, which enables the kompute backend to offload Q4_K_M models.
• Add a --gpu flag to request the model to be offloaded to the GPU.
• When running in a container, bind the server to 0.0.0.0 so the port can be accessed from outside the container.

[slp]

This one supersedes #235

[rhatdan]

LGTM