The HTML spec mandates that id="" be encoded, we follow this in TOC headers but
not in id's specified e.g. with <span>, as a result manually specified links
within a page break if they contain characters that should be encoded.
- <span id="bæt">byte</span>
The inline TOC links will work but not the manually specified backlink
TOC links are encoded as a special case in the parser (see $canonized_headline),
this needs to be put into some general encoding routine in Sanitizer or something.