The previous test suite used a weird "dynamic patching" of a UnitTest
class, where in fact all it needed was a sort of parametrization.

PLEASE NOTE: Some tests have been moved into an /extra folder, they're
testing some special cli flag that we're not really concerned with. The
aim is to include them at some point, but the priority is with the base
html->md tests.

I don't see the point of keeping HTML entities as-is during the
conversion (i.e. not converting & into &). The reason for doing it
that way originally is not convincing:
#109

Empty tables resulted in a spurious `---` to underline a non-existing
table header. Check if there is text before outputting the header marker.

In HTML we have plenty of examples where an emphasis tag contains one or
more <br/> tags. The markdown specification isn't concerned with mapping
html to markdown, so we have to play the assume game. For our purposes,
and to my "common" sense, there's never a case when we want the closing
`**` to be on a different line.

Because converting those html entities leaves a door open to XSS.

Fix the handling of tables, entities and emphasis

(And fix tests accordingly)

Make html2text UNICODE_SNOB by default