improve html / markup validation

Currently, we try to validate the html we get from ipuz files and convert it to be parseable by gmarkup. We don't really follow the spec as a result, as there are subtle differences. To really support this correctly, we must:

• Add a construct-only property as to whether to convert the puzzle to gmarkup or keep it as html
• Convert html entities to unicode characters for gmarkup
• Handle the ipuz:* tag. It's really unclear what this does so we may need more examples to understand it.