There are clear issues remaining with this approach.
The wallabag-given 'start' and 'end' fields do _not_ just point to the
n-th paragraph all the time (like I thought) but actually represent a
beautifulsoup4 like tree descent.
So: `p_start_match = re.match(r"/p\[(\d+)\]", annot["ranges"][0]["start"])`
will fail on any annotation not just at the n-th paragraph.
Instead we should see how we can move this tree into the beautifulsoup4
parser and make use of wallabag already having done the work for us?