v2.4.4

Improve tokenizers (introduced in 2.4.1)

• Improve efficiency of default tokenizer
• Add option to RegExTokenizer to use split_pattern (the pattern that separates tokens, and that will be removed) or token_pattern (the pattern for tokens and that will be retained)
• Make boundary tokens have length zero so that char indexes of text tokens correspond to original text

Full Changelog: 2.4.3...v2.4.4