* Updated README to include installation and usage instructions.
* Added standard open-source tooling, including package testing and requirements.
* Add `serialization` and `beam` modules to the API.
* Added utilities for creating test datasets.
* Added more examples, including a Dirichlet partitioning example.
* Removed `preprocess_fn` arg from `PartitionedDataset.build_group_stream`. Preprocessing should now be applied via `tf.data.Dataset.map`.