Proposal: check model path in --model_path scripts

[jeremyfowers]

Some model scripts, such as https://github.com/onnx/turnkeyml/blob/main/models/transformers/llama2_7b.py, allow the user to pass custom model weights by path. Currently those paths are completely arbitrary, so a user could, for example, accidentally pass LLaMA-2-13b weights to llama2_7b.py, which would give unintended results.

Proposed solution: require some kind of string matching in the model path. For exmaple, all LLaMA-2-7b checkpoints could be required to have "LLaMA-2-7b" or "7b" in the --model_path.

Bonus solution: allow --model_path to be generic with respect to the specific model size. For example, "meta-llama/Llama-2-*b-hf" could load the 7b weights in llama2_7b.py, the 13b weights in llama2_13b.py, etc.

cc @danielholanda