Extend lexer to cover non-ASCII unicode cases

[graydon]

There are a few parts to this:

• Implement the PEP 3131 rule for identifier tokenization: notice when you leave ASCII range, switch to XID_Start/XID_Continue, NFKC-normalize. This is expensive but it'll be a cold path in any ASCII-range input.
• Copy of the rules in rustboot's handling of character and string literals (handle unicode escapes in those contexts, as well as general non-normalizing UTF-8 input).