Plumb checkpoint path for quantizing and tokenizer for 8da4w-gptq (#2478)

Summary:
Pull Request resolved: https://github.com/pytorch/executorch/pull/2478

Quantize (via a transform) during _export_llama was not being provided a checkpoint path, resulting in the same hardcoded path being used.

Similarly, the tokenizer was searched for using dead reckoning from said path, which isn't desired

This diff plumbs both the checkpoint and tokenizer for use in 8da4w-gptq transforms.

Reviewed By: jerryzh168

Differential Revision: D54978447

fbshipit-source-id: b5e9b8e344c6c1c22475d99d5e50e8fba78ce86c