Offer methods for Dataset to cover the most common mechanisms for moving data between partitions

[karlhigley]

The proposed methods would be shuffle_by_keys, sort_by_keys, and group_by_keys. Right now, we only have shuffle_by_keys.

@rjzamora says:

exposing a clear space for documentation is probably the best reason to add it. That documentation should also clarify that these global operations (requiring inter-partition data movement) should be avoided unless absolutely necessary :slightly_smiling_face: