enable loading `universal checkpointing` checkpoint in `DeepSpeedStrategy`

Nobody has claimed this yet.

### Description & Motivation

After I trained a model in some numbers of gpus, say, 8 gpus for a while, It's difficult to load the checkpoint to 16 gpus with optimizer and model states unchanged. The deepspeed has developed the universal checkpointing strategy to solve this problem, but I didn't see the `pytorch-lightning` has this feature.

### Pitch

I want the `pytorch-lightning` could support this feature

### Alternatives

try to add `universal_checkpoint` as a param of `DeepSpeedStrategy` and modify the class refering to `https://www.deepspeed.ai/tutorials/universal-checkpointing/`

### Additional context

_No response_

cc @borda @awaelchli

Start with the DeepSpeedStrategy entry point and the DeepSpeed universal checkpointing tutorial linked in the issue. Trace how model and optimizer checkpoints are currently saved and loaded, then define done as loading a checkpoint created with one GPU count onto a different GPU count while preserving those states.

Written by the indexing model from the issue text.