ramalama client

[ericcurtin]

As a test tool for "ramalama serve" (would also be useful as a follow on feature to connect to remote endpoints like openai, perplexity, etc.).

Should behave identically to "ramalama run", except it does not inference within the client process, it should send a request to the server process and stream the results back.

Can exist here as python or, it could also be written in C++ and eventually contributed back to llama.cpp upstream.

[ericcurtin]

Probably makes more sense as a ramalama-only tool in python as the official openai library is in python:

https://github.com/openai/openai-python

[ericcurtin]

Since this will rely on python library outside the standard python library, by default we should execute this inside the container

[ericcurtin]

This will also help us test vllm, since vllm only operates as a server