HTMLParser, which is used by extract_attributes, already unescapes attribute values with HTMLParser.unescape. They shouldn't be unescaped again, to there may be parsing errors. Ref: #11219, #11522