Fix sanitizer with libxml2 >= 2.12.0
Somehow with newer libxml2, <?xml encoding="UTF-8"> no longer enforces UTF-8. Instead, non-ASCII contents are treated as ISO-8859-1 and get broken.
For example, <p>中文</p> becomes <p>中文</p> (should be <p>中文</p>).
Switching to another trick fixes the issue, and the new trick still works with older libxml2 (tested 2.11.5).
As a side note, DOMDocument::loadHTML uses HTMLParser in libxml2.