Improve dart-lang/markdowndart-lang/markdown#51.

First, put a try/catch around attempting to parse Markdown.
Failure to parse should count as a negative test case, but
should not abort testing altogether.

Second, implement a primitive HTML comparison function that
is not sensitive to attribute order, meaningless whitespace,
etc.

The second change alone boosts CommonMark score from 232/599
to 383/599.