This fixes a performance regression when mentions are being read
repeatedly from the API and we have to re-render all the notes to
parse the HTML