Add nvtext::tokenize_with_vocabulary API

[davidwendt]

Description

Adds tokenize with vocabulary APIs to libcudf.

struct tokenize_vocabulary{ ... };

std::unique_ptr<tokenize_vocabulary> load_vocabulary(
  cudf::strings_column_view const& input,
  rmm::cuda_stream_view stream,
  rmm::mr::device_memory_resource* mr);

std::unique_ptr<cudf::column> tokenize_with_vocabulary(
  cudf::strings_column_view const& input,
  tokenize_vocabulary const& vocabulary,
  cudf::string_scalar const& delimiter,
  cudf::size_type default_id,
  rmm::cuda_stream_view stream,
  rmm::mr::device_memory_resource* mr);

Returns an integer lists column replacing individual tokens as resolved from the input using delimiter with id values which are the row indices of the input vocabulary column.
If a token is not found in the vocabulary it is assigned default_id.
The vocabulary can be loaded once using the nvtext::load_vocabulary() API and then used in repeated calls to nvtext::tokenize_with_vocabulary() with different input columns.

Python interface is new class TokenizeVocabulary which can be used like the following:

>>> import cudf
>>> from cudf.core.tokenize_vocabulary import TokenizeVocabulary
>>> words = cudf.Series( ['brown', 'the', 'dog', 'jumps'] )
>>> vocab = TokenizeVocabulary(words)
>>> s = cudf.Series( ['the brown dog jumps over the brown cat'] )
>>> print(vocab(s))
0    [1, 0, 2, 3, -1, 1, 0, -1]
dtype: list

Checklist

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

[robertmaynard]

Approving CMake changes

[nvdbaranec]

Looks good. Just a couple of tiny things.

[bdice]

Feedback attached. The cuco static map is truly awesome. 👍

[bdice]

One suggestion, otherwise LGTM.

[davidwendt]

/merge