Remove dependency on llama_models

[raghotham]

🚀 Describe the new functionality needed

Copy over tokenizer and datatypes from llama-model

💡 Why is this needed? What if we don't build it?

Right now we are depending on meta-llama/llama-models in llama-stack. This causes us to inherit that repo's license. We would like to keep llama-stack to be purely MIT license.

Other thoughts

No response