Normalize away unnecessary attributes in data-mw.html too
"about" attributes on elements in data-mw.html are another source of
indeterminacy, as noted in T93715, but were being missed because of the
JSON encoding not matching the regexp.
The new errors on,
node bin/roundtrip-test.js --domain en.wikipedia.org "Nicolas Iljine"
were being classified as semantic when the numbers didn't line up.