Using SAX style parsing instead of XPath searches

* SAX requires less memory
* Performance doesn't differ apparently
* Also: improved error handling
* Also: moved from stdlib to stream conversions