Detect any that contains a "-[a-z]" (i.e. a dash and a lowercase letter.
So far it was taking "digit*-digit*-anything*", which is clearly not
enough.
Also since we're already parsing the svg file in some cases, use that
parsing to detect the interesting ids.