Add BytePairEncoder class to cuDF

[davidwendt]

Description

Adds a new BytePairEncoding class to cuDF

>>> import cudf
>>> from cudf.core.byte_pair_encoding import BytePairEncoder
>>> mps = cudf.read_text('merges.txt', delimiter='\n', strip_delimiters=True)
>>> bpe = BytePairEncoder(mps)
>>> str_series = cudf.Series(['This is a sentence', 'thisisit'])
>>> bpe(str_series)
0    This is a sent ence
1             this is it
dtype: object

This class wraps the existing nvtext::byte_pair_encoding APIs to load the merge-pairs data and encode a column of strings.

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[bdice]

Approving. Requesting one small rename and one docstring summary.

[davidwendt]

/merge