This adds support for bytes literals (b'...') to tokenize.py, and
removes support for unicode literals (u'...').