`.shuffle` throwing error `ValueError: Protocol not known: parent`

### Describe the bug

```
---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
Cell In [16], line 1
----> 1 train_dataset = train_dataset.shuffle()

File /opt/conda/envs/pytorch/lib/python3.9/site-packages/datasets/arrow_dataset.py:551, in transmit_format..wrapper(*args, **kwargs)
544 self_format = {
545 "type": self._format_type,
546 "format_kwargs": self._format_kwargs,
547 "columns": self._format_columns,
548 "output_all_columns": self._output_all_columns,
549 }
550 # apply actual function
--> 551 out: Union["Dataset", "DatasetDict"] = func(self, *args, **kwargs)
552 datasets: List["Dataset"] = list(out.values()) if isinstance(out, dict) else [out]
553 # re-apply format to the output

File /opt/conda/envs/pytorch/lib/python3.9/site-packages/datasets/fingerprint.py:480, in fingerprint_transform.._fingerprint..wrapper(*args, **kwargs)
476 validate_fingerprint(kwargs[fingerprint_name])
478 # Call actual function
--> 480 out = func(self, *args, **kwargs)
482 # Update fingerprint of in-place transforms + update in-place history of transforms
484 if inplace: # update after calling func so that the fingerprint doesn't change if the function fails

File /opt/conda/envs/pytorch/lib/python3.9/site-packages/datasets/arrow_dataset.py:3616, in Dataset.shuffle(self, seed, generator, keep_in_memory, load_from_cache_file, indices_cache_file_name, writer_batch_size, new_fingerprint)
3610 return self._new_dataset_with_indices(
3611 fingerprint=new_fingerprint, indices_cache_file_name=indices_cache_file_name
3612 )
3614 permutation = generator.permutation(len(self))
-> 3616 return self.select(
3617 indices=permutation,
3618 keep_in_memory=keep_in_memory,
3619 indices_cache_file_name=indices_cache_file_name if not keep_in_memory else None,
3620 writer_batch_size=writer_batch_size,
3621 new_fingerprint=new_fingerprint,
3622 )

File /opt/conda/envs/pytorch/lib/python3.9/site-packages/datasets/arrow_dataset.py:551, in transmit_format..wrapper(*args, **kwargs)
544 self_format = {
545 "type": self._format_type,
546 "format_kwargs": self._format_kwargs,
547 "columns": self._format_columns,
548 "output_all_columns": self._output_all_columns,
549 }
550 # apply actual function
--> 551 out: Union["Dataset", "DatasetDict"] = func(self, *args, **kwargs)
552 datasets: List["Dataset"] = list(out.values()) if isinstance(out, dict) else [out]
553 # re-apply format to the output

File /opt/conda/envs/pytorch/lib/python3.9/site-packages/datasets/fingerprint.py:480, in fingerprint_transform.._fingerprint..wrapper(*args, **kwargs)
476 validate_fingerprint(kwargs[fingerprint_name])
478 # Call actual function
--> 480 out = func(self, *args, **kwargs)
482 # Update fingerprint of in-place transforms + update in-place history of transforms
484 if inplace: # update after calling func so that the fingerprint doesn't change if the function fails

File /opt/conda/envs/pytorch/lib/python3.9/site-packages/datasets/arrow_dataset.py:3266, in Dataset.select(self, indices, keep_in_memory, indices_cache_file_name, writer_batch_size, new_fingerprint)
3263 return self._select_contiguous(start, length, new_fingerprint=new_fingerprint)
3265 # If not contiguous, we need to create a new indices mapping
-> 3266 return self._select_with_indices_mapping(
3267 indices,
3268 keep_in_memory=keep_in_memory,
3269 indices_cache_file_name=indices_cache_file_name,
3270 writer_batch_size=writer_batch_size,
3271 new_fingerprint=new_fingerprint,
3272 )

File /opt/conda/envs/pytorch/lib/python3.9/site-packages/datasets/arrow_dataset.py:551, in transmit_format..wrapper(*args, **kwargs)
544 self_format = {
545 "type": self._format_type,
546 "format_kwargs": self._format_kwargs,
547 "columns": self._format_columns,
548 "output_all_columns": self._output_all_columns,
549 }
550 # apply actual function
--> 551 out: Union["Dataset", "DatasetDict"] = func(self, *args, **kwargs)
552 datasets: List["Dataset"] = list(out.values()) if isinstance(out, dict) else [out]
553 # re-apply format to the output

File /opt/conda/envs/pytorch/lib/python3.9/site-packages/datasets/fingerprint.py:480, in fingerprint_transform.._fingerprint..wrapper(*args, **kwargs)
476 validate_fingerprint(kwargs[fingerprint_name])
478 # Call actual function
--> 480 out = func(self, *args, **kwargs)
482 # Update fingerprint of in-place transforms + update in-place history of transforms
484 if inplace: # update after calling func so that the fingerprint doesn't change if the function fails

File /opt/conda/envs/pytorch/lib/python3.9/site-packages/datasets/arrow_dataset.py:3389, in Dataset._select_with_indices_mapping(self, indices, keep_in_memory, indices_cache_file_name, writer_batch_size, new_fingerprint)
3387 logger.info(f"Caching indices mapping at {indices_cache_file_name}")
3388 tmp_file = tempfile.NamedTemporaryFile("wb", dir=os.path.dirname(indices_cache_file_name), delete=False)
-> 3389 writer = ArrowWriter(
3390 path=tmp_file.name, writer_batch_size=writer_batch_size, fingerprint=new_fingerprint, unit="indices"
3391 )
3393 indices = indices if isinstance(indices, list) else list(indices)
3395 size = len(self)

File /opt/conda/envs/pytorch/lib/python3.9/site-packages/datasets/arrow_writer.py:315, in ArrowWriter.__init__(self, schema, features, path, stream, fingerprint, writer_batch_size, hash_salt, check_duplicates, disable_nullable, update_features, with_metadata, unit, embed_local_files, storage_options)
312 self._disable_nullable = disable_nullable
314 if stream is None:
--> 315 fs_token_paths = fsspec.get_fs_token_paths(path, storage_options=storage_options)
316 self._fs: fsspec.AbstractFileSystem = fs_token_paths[0]
317 self._path = (
318 fs_token_paths[2][0]
319 if not is_remote_filesystem(self._fs)
320 else self._fs.unstrip_protocol(fs_token_paths[2][0])
321 )

File /opt/conda/envs/pytorch/lib/python3.9/site-packages/fsspec/core.py:593, in get_fs_token_paths(urlpath, mode, num, name_function, storage_options, protocol, expand)
591 else:
592 urlpath = stringify_path(urlpath)
--> 593 chain = _un_chain(urlpath, storage_options or {})
594 if len(chain) > 1:
595 inkwargs = {}

File /opt/conda/envs/pytorch/lib/python3.9/site-packages/fsspec/core.py:330, in _un_chain(path, kwargs)
328 for bit in reversed(bits):
329 protocol = split_protocol(bit)[0] or "file"
--> 330 cls = get_filesystem_class(protocol)
331 extra_kwargs = cls._get_kwargs_from_urls(bit)
332 kws = kwargs.get(protocol, {})

File /opt/conda/envs/pytorch/lib/python3.9/site-packages/fsspec/registry.py:240, in get_filesystem_class(protocol)
238 if protocol not in registry:
239 if protocol not in known_implementations:
--> 240 raise ValueError("Protocol not known: %s" % protocol)
241 bit = known_implementations[protocol]
242 try:

ValueError: Protocol not known: parent

```

This is what the `train_dataset` object looks like
```
Dataset({
features: ['label', 'input_ids', 'attention_mask'],
num_rows: 364166
})
```

### Steps to reproduce the bug

The `train_dataset` obj is created by concatenating two datasets
And then shuffle is called, but it throws the mentioned error.

### Expected behavior

Should shuffle the dataset properly.

### Environment info

- `datasets` version: 2.6.1
- Platform: Linux-5.15.0-1022-aws-x86_64-with-glibc2.31
- Python version: 3.9.13
- PyArrow version: 10.0.0
- Pandas version: 1.4.4