Add Dataset.pipe method, based on pandas.DataFrame.pipe.

[copybara-service]

Add Dataset.pipe method, based on pandas.DataFrame.pipe.

pipe is convenient because it allows for using method chaining syntax in an extensible fashion, with transformations that are not built-in methods on Dataset.

For example, consider shuffling a dataset in windows. It would be convenient if we could write something like:

ds = (
    dataset.MapDataset.range(400)
    .window_shuffle(window_size=10, seed=42)
    .batch(16)
    .repeat()
)

Unfortunately this doesn't work, because there is no window_shuffle() method. Instead you would need to write something like:

ds = (
    shuffle.WindowShuffleMapDataset(
        dataset.MapDataset.range(400),
        window_size=10,
        seed=42,
    )
    .batch(16)
    .repeat()
)

Readability suffers here, because the shuffle transformation comes out of order.

Instead, pipe lets us write something like, keeping transformations in the order in which they are applied:

ds = (
    dataset.MapDataset.range(400)
    .pipe(
        shuffle.WindowShuffleMapDataset,
        window_size=10,
        seed=42,
    )
    .batch(16)
    .repeat()
)