Segment a word with greedy longest-match WordPiece tokenization against a supplied subword vocabulary.
