Add image diff tests

[jan-xyz]

Test the output from the visualizers against expected images.

When we have that we can also submit these pictures for calculation and make sure that the probabilities are as expected.

That means:

• collect some sample sumbers for the MNIST model testing
• compare output images against expected images
• compare output probabilieties against expected images

[jan-xyz]

stack overflow idea: https://stackoverflow.com/questions/29261009/carrying-out-unit-testing-in-python-on-a-method-that-implements-imagedraw

[jan-xyz]

I think this can be closed as it was part of PR #41