Automated testing

[mdraw]

The "tests" in this repository are currently not useful for automated testing. They are more like interactive demonstrations that show how you can use parts of the internal ELEKTRONN2 API directly.
Implementing actual unit and integration tests that can be automatically run (in CI) by py.test or similar would be an important step for facilitating further development and making collaboration easier and safer.

IMO our first focus should be on simple high-level tests that prove that basic task like building/saving/loading models, running minimal CNNs etc. work like intended, so we would immediately notice if a change in the codebase accidentally breaks basic functionality.
When designing tests, we have to keep in mind that (almost) all of them should be able to run on CI servers, so we (currently) can't rely on GPUs and should keep the computational cost and time consumption of tests minimal.