fixed issue #20065

[rohan-devbuilds]

What does this PR do?

Fixes #20065

Adds support for DeepSpeed's Universal Checkpointing feature in DeepSpeedStrategy.

Previously, if you trained a model on N GPUs and tried to resume training on a
different number of GPUs (e.g. 8 → 16), the checkpoint would be incompatible and
optimizer/model states would be lost. DeepSpeed provides a built-in solution called
Universal Checkpointing, but Lightning had no way to enable it.

This PR adds a load_full_weights: bool = False parameter to DeepSpeedStrategy.
When set to True, it injects {"checkpoint": {"load_universal": True}} into the
DeepSpeed config, enabling seamless checkpoint loading across different GPU counts.

Example usage:

strategy = DeepSpeedStrategy(load_full_weights=True)
trainer = Trainer(strategy=strategy)
trainer.fit(model, ckpt_path="path/to/universal_checkpoint")
` `` 

**Changes made:**
- Added `load_full_weights: bool = False` parameter to `DeepSpeedStrategy.__init__`
- Injected `checkpoint.load_universal` into the DeepSpeed config when the flag is `True`
- Updated docstring to document the new parameter
- Added unit tests for both the enabled and default (disabled) cases


  Before submitting

- Was this **discussed/agreed** via a GitHub issue? (not for typos and docs)
  - Yes — see #20065
- [x] Did you read the [contributor guideline](https://github.com/Lightning-AI/pytorch-lightning/blob/master/.github/CONTRIBUTING.md), **Pull Request** section?
- [x] Did you make sure your **PR does only one thing**, instead of bundling different changes together?
- Did you make sure to **update the documentation** with your changes? (if necessary)
  - Yes — updated the `__init__` docstring in `DeepSpeedStrategy`
- Did you write any **new necessary tests**? (not for typos and docs)
  - Yes — added tests verifying config is correctly set when `load_full_weights=True` and clean when `False`
- [x] Did you verify new and **existing tests pass** locally with your changes?
- Did you list all the **breaking changes** introduced by this pull request?
  - No breaking changes. New parameter defaults to `False`, preserving existing behavior.
- Did you **update the CHANGELOG**?
  - Yes — added entry under unreleased section



## PR review

Anyone in the community is welcome to review the PR.
Before you start reviewing, make sure you have read the
[review guidelines](https://github.com/Lightning-AI/lightning/wiki/Review-guidelines).


  Reviewer checklist

- [x] Is this pull request ready for review?
- [x] Check that all items from **Before submitting** are resolved
- [x] Title is self-explanatory and description concisely explains the PR
- [ ] Add labels and milestones (and optionally projects) to the PR so it can be classified

---

📚 Documentation preview 📚: https://pytorch-lightning--21683.org.readthedocs.build/en/21683/