Add a llama model completely written in Rust behind a feature

[Luni-4]

Adding the Burn llama model behind a feature would avoid having llama.cpp installed on the system.

I do not know if it is feasible because I have not checked the code thoroughly, this is just an idea. Feel free to close this issue if this proposal is deemed out of line.

[hecrj]

I think we should eventually make the server address configurable; so users can target anything that implements the OpenAI completion API.

[Luni-4]

That seems an acceptable solution also, thank you!