Add LLM continual pre-training example

[KevinMusgrave]

Using a recent model like llama3 or mistral v0.3