'huggingface:mnist' dataset not generating

Short description
Loading the huggingface:mnist gives ArributeError

Environment information

• Operating System: Windows 10

• Python version: 3.8

• tensorflow-datasets/tfds-nightly version: 4.2.0+nightly

• tensorflow/tf-nightly version: 2.4.0

• Does the issue still exists with the last tfds-nightly package (pip install --upgrade tfds-nightly) ? Yes

Reproduction instructions

>>> nik = tfds.load('huggingface:mnist')
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "f:\tfds\datasets\tensorflow_datasets\core\load.py", line 325, in load
    dbuilder = builder(name, data_dir=data_dir, try_gcs=try_gcs, **builder_kwargs)
  File "f:\tfds\datasets\tensorflow_datasets\core\load.py", line 172, in builder
    return cls(**builder_kwargs)  # pytype: disable=not-instantiable
  File "f:\tfds\datasets\tensorflow_datasets\core\dataset_builder.py", line 947, in __init__
    super().__init__(**kwargs)
  File "f:\tfds\datasets\tensorflow_datasets\core\dataset_builder.py", line 182, in __init__
    self.info.initialize_from_bucket()
  File "f:\tfds\datasets\tensorflow_datasets\core\utils\py_utils.py", line 149, in __get__
    cached = self.fget(obj)  # pytype: disable=attribute-error
  File "f:\tfds\datasets\tensorflow_datasets\core\community\huggingface_wrapper.py", line 124, in info
    info = self._info()
  File "C:\Users\nikhil\.cache\tensorflow_datasets\modules\tfds_community\huggingface\mnist\cf39692eaead13bb56fce5bcd2a1b230769d6801ce84feca3620e4201fdb8706\mnist.py", line 68, in _info
    "image": datasets.Array2D(shape=(28, 28), dtype="uint8"),
AttributeError: Failed to construct dataset huggingface:mnist: module 'datasets' has no attribute 'Array2D'

Expected behavior
Since, the datasets library has a well defined Array2D method, so there souldn't have been any error and the dataset should have been generated seamlessly.

Additional context
This issue might be part of a larger problem with interfacing huggingface datasets with TFDS as there seems to be unusual errors with other HF datasets in TFDS as well like huggingface:xnli etc.

Start with tensorflow_datasets/core/community/huggingface_wrapper.py and the generated huggingface:mnist module shown in the traceback, then reproduce the failure with tfds.load('huggingface:mnist'). Check how the installed datasets library exposes Array2D and related Hugging Face dataset types; done means the MNIST dataset loads without the reported AttributeError and the related compatibility behavior is understood.