feat: find a solution to load the dataset

While testing the real data extraction, I encountered a new problem: the websites descriptions are rarely present in the `metadata_website_desc` column. Therefore, the datasets library cannot load such a dataset by having to guessing the feature type. It has to know them beforehand.