Lexer now differentiates identifiers and keywords.

With that generalized the regex token detection - a division operator
can never follow a keyword, but regex literal can.