Streamline/better documentation for torchtune -> transformers workflow

[SalmanMohammadi]

We should make it easier for people to load models trained from torchtune into transformers (we could also document using huggingface-cli for upload too?). This could be better documentation i.e. highlighting the necessary steps to load a torchtune-hf checkpoint into a transformers model (just a case of renaming the final checkpoint?), or doing some plumbing ourselves, like saving final model checkpoints with filenames that are readily consumable by from_pretrained

ref #1381 #832 #878 #1122