Fix issue with transformers library huggingface

[alexcbb]

Description

I was trying to modifiy a RLDS dataset built on top of TFDS following this repo : https://github.com/kpertsch/rlds_dataset_builder
I needed to extract some features from images with models from the transformers library of HuggingFace but was facing an issue during the import : raise ValueError('{}.__spec__ is None'.format(name)) ValueError: datasets.__spec__ is None

And more specifically this one : transformers/utils/import_utils.py", line 120, in <module> _datasets_available = _is_package_available("datasets")

It verifies if datasets (the HF library) is available by looking at the __spec__ attribute. As tfds is overwritting datasets by a mock, it does not create the attribute, which causes the issue.

In the PR I fixed the issue by simply creating the needed attribute in the Mock in order to solve the problem.

Here are the versions of the libraries involved in the problem :
tensorflow-datasets 4.9.3
transformers 4.50.0.dev0