In particular, saving large checkpoints >2GB will hit "Requested array size exceeds VM limit" error when writing to HDFS through pyarrow:

```
  File "/usr/local/lib/python3.9/dist-packages/pytorch_lightning/strategies/strategy.py", line 466, in save_checkpoint
    self.checkpoint_io.save_checkpoint(checkpoint, filepath, storage_options=storage_options)
  File "/usr/local/lib/python3.9/dist-packages/lightning_fabric/plugins/io/torch_io.py", line 54, in save_checkpoint
    _atomic_save(checkpoint, path)
  File "/usr/local/lib/python3.9/dist-packages/lightning_fabric/utilities/cloud_io.py", line 72, in _atomic_save
    f.write(bytesbuffer.getvalue())
  File "pyarrow/io.pxi", line 359, in pyarrow.lib.NativeFile.write
  File "pyarrow/error.pxi", line 113, in pyarrow.lib.check_status
OSError: [Errno 12] HDFS Write failed. Detail: [errno 12] Cannot allocate memory
```