* refactor: separate token and message reading assistance modes * chore: apply same token styling to HTML formatted messages * chore: don't wait for lemma responses before showing reading assistance content
