This change adds the \u notation for double quoted strings and regular
expressions.  It does not implement the \u notation for character literals.
Both the \uNNNN and \u{NNNN} notations are supported.

\uNNNN is implemented by emitting equivalent UTF-8; that is, "\u4000" is
equivalent to "\xE4\x80\x80".

Unlike CRuby, the \u{NNNN} notation allows only one character per pair of
braces; I see no way to lift this restriction without remodeling the parser.