refactor the lexer

[mgerhold]

At the moment the lexer uses one huge function (tokenize) to tokenize the code. The code could become cleaner, if this could be refactored into several functions. For this some kind of LexerState-struct would be necessary.