Specify columns when reading files with DocumentDataset

[sarahyurick]

Closes #180.

All of these will now work:

(1) Pandas and cuDF read_json do not support a columns parameter, so we read in the entire DataFrame and then remove unwanted columns behind the scenes.

dataset = DocumentDataset.read_json(dataset_path, columns=["col1", "col2"])

(2) Pandas and cuDF read_parquet both support a columns parameter, so we are able to take advantage of this functionality.

dataset = DocumentDataset.read_parquet(dataset_path, columns=["col1", "col2"])

(3) Pandas read_pickle (there is no cuDF read_pickle) does not support a columns parameter, so we read in the entire DataFrame and then remove unwanted columns behind the scenes.

dataset = DocumentDataset.read_pickle(dataset_path, columns=["col1", "col2"])

(4) Following cudf.read_json, you can specify dtype and prune_columns=True to only return the columns mentioned in the dtype argument. Note that Pandas does not support prune_columns.

dataset = DocumentDataset.read_json(
    dataset_path,
    dtype={"col1": str, "col2": str},
    prune_columns=True,
    backend="cudf",
)

[ryantwolf]

One minor thing, then we should be good.

[ryantwolf]

Good with me, thanks!