[Train] LightningTrainer enable checkpoint full dict with FSDP strategy

[woshiyyya]

Why are these changes needed?

In PyTorch Lightning 2.0, when using the FSDP strategy, the checkpoint does not include the model state dictionary. This PR aims to add a model unsharding logic, which gathers all model shard from workers to rank0 CPU memory.

This is a pre-request PR for FSDP LLM finetune example: #34990

Related issue number

Checks

• I've signed off every commit(by using the -s flag, i.e., git commit -s) in this PR.
• I've run scripts/format.sh to lint the changes in this PR.
• I've included any doc changes needed for https://docs.ray.io/en/master/.
  • I've added any new APIs to the API Reference. For example, if I added a
    method in Tune, I've added it in doc/source/tune/api/ under the
    corresponding .rst file.
• I've made sure the tests are passing. Note that there might be a few flaky tests, see the recent failures at https://flakey-tests.ray.io/
• Testing Strategy
  • Unit tests
  • Release tests
  • This PR is not tested :(

[gjoliver]

A quick question.