Improve test coverage of marbles \times unittest

`marbles` itself is fairly comprehensively tested, in terms of lines and also features, but we don't very well test how marbles interacts with all of the features of `unittest`. For example, some of the assertions we're wrapping in `marbles` may behave strangely depending on how we've wrapped them (we've seen problems before where things like `assertMultiLineEqual` don't faithfully pass the `msg` argument to all assertions they then make, which broke our annotations, but only when that method got called with something that wasn't a `str`).

We should try to test `marbles` against all assertions in `unittest`, possibly by running `unittest`'s own test harnesses. We may also want to test interactions between `marbles` and other features of `unittest`, but let's start there.