Move tokenizer information into pte to reduce ExecuTorch runner args

[Jack-Khuu]

🚀 The feature, motivation and pitch

After an ExecuTorch model is exported to a pte, tokenization information must be passed in as an arg (-l <#>) to the runner. This can be avoided by writing this information into the pte file itself since the tokenizer is known at export time (sentencepiece => 2, tiktoken =>3). Tokenization information can be stored during export as a constant_method.

For example: https://github.com/pytorch/torchchat?tab=readme-ov-file#deploy-and-run-on-android

cmake-out/et_run llama3.1.pte -z `python3 torchchat.py where llama3.1`/tokenizer.model -l 3 -i "Once upon a time"

---

Task:

1. Update ExecuTorch exporting to save tokenization information in the pte artifact
2. Update the ExecuTorch runner to read the newly saved metadata

For a similar optimization made for aoti: #1159.
See #1439 for conversation/more context

Alternatives

Continue to pass tokenizer arguments to the runner

Additional context

No response

RFC (Optional)

No response