Add evals support for 3.x meta Llama eval datasets in lm-evaluation-harness

[tstescoTT]

Datasets are configured with yaml files per model / task with a dataset on HF e.g. https://huggingface.co/datasets/meta-llama/Llama-3.2-1B-Instruct-evals

See meta team documentation at https://github.com/meta-llama/llama-cookbook/tree/ef99b32e48c8019aa64bd562f8138b8b84049d07/end-to-end-use-cases/benchmarks/llm_eval_harness/meta_eval