Pickle dataset features file #908

Features files can grow very large, which makes reading them slow. As an example, have a look at https://www.openml.org/api/v1/data/features/41147

As a solution, I suggest pickling the feature file to load the data structure from disk.