The `diceware` run in travis was meant as a packaging test. It was
supposed to fail if not all data files were available in
installs. Unfortunately it did not fail in cases where `pip` *did*
fail.

Furthermore we want to detect such packaging problems in regular
tests, before pushing changes.

As we do not install extra packages we do not have to wait for
`apt-get update`.

It is more efficient to run only one test for Python 2.x and one for
Python 3.x. The full set of tests, including all supported versions
listed in `setup.py` should be run locally at least before
releases. Then we also do coverage tests, pep8 and pyflakes.

The 'run diceware' test should remain until we have some other
regression test for bug #1.