Add option to avoid automatic consumption of new-lines by lexemes about megaparsec

[mrkkrp]

The problem is solved in ded90fd. I've chosen to give more control to the user. This also make it more difficult to write parsers, but eliminates the need to rewrite entire Text.Megaparsec.Lexer module every time user needs to do something non-standard.

More information is in commit message. I expect that makeLexer will be a lot simpler and user will need to do some parsing manually. We can also provide helper functions to construct parser that parses block comments for example, and then use the result in definition of language. It's better than the old scheme where all the logic is locked inside of makeTokenParser.