Allow low-level selection of shard iteration strategy. https://github.com/google/sedpack/pull/198 is the first step for `as_numpy_iterator`. Then also other functions `as_numpy_*` and `as_tfdataset`. Then support for advanced iteration strategies. Then code cleanup of `dataset_iteration.py`.
