Making llama prompts configurable

[ghost]

Basically to allow more varity in the response and also change the prompts based on the model's supported format e.g. alpaca vs vicuna

[choombaa]

Should we have source-controlled "personas" for reference?

[yacineMTB]

ya