Lexer always parsed numbers as unsigned, but parser handled them as signed and grammar contained many unnecessary checks for negativity.
