Make Llava to be configurable so that you can swap text model

[mergennachin]

🚀 The feature, motivation and pitch

Current Llava uses Llama 2 7B as the pretrained text model. https://github.com/pytorch/executorch/blob/main/examples/models/llava/README.md

The latest quantized Llama 1B/3B are good in terms of accuracy and size. Let's make it swappable with these models.

Alternatives

No response

Additional context

No response

RFC (Optional)

No response