fix(cli): allow passing n_ctx=0 to openAI API server CLI arguments

[K-Mistele]

What?

This PR updates the OpenAI API server's command line arguments to allow passing --n_ctx 0 as a command-line argument. Currently, --n_ctx parameter has a minimum of 1: in llama_cpp/server/settings.py you can see the following context CLI parameter configuration: n_ctx: int = Field(default=2048, ge=1, description="The context size.")

Why?

As of [email protected], per #1015 by @DanieleMorotti, passing n_ctx=0 to the LLama class in llama_cpp/llama.py automatically sets the n_ctx to the model's n_ctx_train paramter from KV, and also updates the model's n_batch to min(n_ctx, n_batch). This is intentional to allow llama-cpp-python to infer the context size from the GGUF model file's KV parameters.

However, when this change was made, the OpenAI API server's CLI argument configuration was not updated, so the minimum value for the option remained at 1 - making the patch in #1015 unavailable to users of the OpenAI API serve.r

[abetlen]

@K-Mistele thank you!