I have written a tonne of tests that compare against A&B's original C++.
Until I can distribute the `anders_briegel` package which implements
that original code, we now skip those tests if we see an ImportError.

TODO: seems a little over-zealous right now ...

We fail when running against Anders and Briegel simply because we use a
different CZ table

But it is not symmetric. I wonder why. Can run with either for now.

Using `combinations` instead of `combinations_with_replacement` was
leading to entries of the form [ ... , i, i] being excluded from the CZ
table.

At least we have infinite-size graphs now