Sometimes, tables have row-level headings as well as column level
headings.

To be accessible, these need to be marked up as `<th>` tags. Otherwise
users of screen readers and other assistive technology may not be able
to work out that they're table headings rather than ordinary table
cells.

This allows us to handle the more complex case where table headings
contain other markup (e.g. the heading might have an `<em>` tag, as per
the test).

It feels less hacky than using a regex, but it's quite a bit more code
so I'm not 100% sure whether it's an improvement.