Allow passing custom reader/writer in _distributed_checkpoint_save and _distributed_checkpoint_load.

Nobody has claimed this yet.

### Description & Motivation

[distributed_checkpoint_save](https://github.com/Lightning-AI/pytorch-lightning/blob/master/src/lightning/fabric/strategies/fsdp.py#L861) for sharded dict initializes [torch.distributed.checkpoint.FileSystemWriter](https://github.com/Lightning-AI/pytorch-lightning/blob/master/src/lightning/fabric/strategies/fsdp.py#L876)

Similarly [distributed_checkpoint_load](https://github.com/Lightning-AI/pytorch-lightning/blob/master/src/lightning/fabric/strategies/fsdp.py#L880) and [load_checkpoint](https://github.com/Lightning-AI/pytorch-lightning/blob/master/src/lightning/pytorch/strategies/fsdp.py#L613) initializes torch.distributed.checkpoint.FileSystemReader.

If PyTorch-lightining could provide an integration point to plugin custom readers/writers it would be provide user with freedom to read/write from whatever storage they want.

### Pitch

I want to use pytorch lightning to read/write from GCS bucket using FSDP & sharded checkpoint.
This can be done easily in single node checkpointing since it follows checkpoint IO interface and easily allows to read/write checkpoints from anywhere we want.

However this is not the case for sharded checkpoints in multi node scenario. Having the ability to plugin custom read/writer will help us read/write from storage we want.

### Alternatives

_No response_

### Additional context

_No response_

cc @lantiga @borda

Read the distributed_checkpoint_save and distributed_checkpoint_load entry points in src/lightning/fabric/strategies/fsdp.py, along with load_checkpoint in src/lightning/pytorch/strategies/fsdp.py, and inspect where FileSystemWriter and FileSystemReader are initialized. Done means users can provide custom reader and writer implementations for sharded, multi-node FSDP checkpoint saving and loading.

Written by the indexing model from the issue text.