Support TP for save_pretrained()

[amd-xiaoyu12]

What does this PR do?

When a model is quantized using TP, the save_pretrained() need to support TP when quantization is done and export to model to safetensor file.

The modification will combine the DTensor to full tensor in process which rank = 0 and export the result, for process with other rank number, the exporting task will be skipped.

Fixes # (issue)

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

[BowenBao]

it's likely a problem that this changes the state of the model and makes it invalid after calling save_pretrained.

The in memory model should ideally be the same before and after calling save_pretrained.

[Rocketknight1]

cc @ArthurZucker @Cyrilvallez

[ArthurZucker]

Hey! We have #37919 which should handle this as well! It is a bit more involved for some models, like LLama4, so let's focus on the other PR!
But thanks a lot for the contribution!