now that we have multiple test delegations, it would make sense to introduce
some generalization.

(A DS normally doesn't exist there.)

They might fail when it is implemented, the RRSigs and other details
might need some correction then.

As it is tightly coupled with the previous try statement.

The server is a separate entity from the developer's collective mind,
which, at last for the time being, is not considered sentient. Therefore
the word "we" is not appropriate and the server should be described in
third person.

addNXRRsetDenial() -> addNXRRsetProof()

also added a few comments to test data

removed before releasing the wrapper code.  Compiler will detect if
any of them is forgotten to be cleaned up.

(there doesn't seem to be need for updating the python doc)

as suggested in review.  hid the concrete derived class in the implementation.
adjusted tests accordingly.

It happens to work already.

If it is not one of the "known to be OK" exceptions, we abort.

we should either/both consider revisiting the interface so we avoid the
issue as a result or require higher version of boost.  but for now this
should be okay.