ZPublisher should be smart about encoding errors

[d-maurer]

Modern browsers internally use unicode. Only at system boundaries they employ an encoding to map from/to the internal representation. Following HTML 5, they use character references to represent characters not encodable by the output encoding.

ZPublisher should follow a similar approach for HTML and XML responses (i.e. use
"xmlcharrefreplace" rather than "replace" as encoding error handling). For HTML and XML responses, character references are at most places correctly handled. There are places where this is not the case (e.g. in CDATA sections) but at those places a character reference is not more bad than a replacement character. Maybe, this argument is even correct for all responses (thus that xmlcharrefreplace could be used independent of the reponse type).