There are incompatible changes in some Python modules that make the
tests fail. As it is a new test, allow the failure until this is
fixed.

This way we can also run Python 3 tests in parallel without requiring
more resources in total.

The unit tests are run together with the other tests, so we don't need
to run them separately anymore.

We're not running stretch anymore. Also, some of the six functions are
not available on stretch.

Avoids failure if the py3 job is manually started.

At least `bzip2` and `xz-utils` were not installed for the Python 3 test.

This makes this generated diff a bit smaller.

Based on patches by Niels Thykier to add coverage to britney.

Whilst flake8 will always be Python 2.x in the specified "debian:9" image, we
can make this future-proof and less confusing to anyone reading this and
attempting to reproduce the output on newer distribution where flake8 will
assume Python 3 (and thus report syntax errors against dak's Python 2.x code).