Unit tests using a pre-trained mnist model

[anagainaru]

For now I would start with tests for each separate function:

• all methods for drift detection (making sure the drift is detected when we want and it's not detected when it shouldn't)
• continual learning (that the accuracy increases by how much we expect for an example)
• the logger to make sure what we log is correct

This can be separate in 3 parts (when you take ownership of this comment as to which part you are doing)

[ScSteffen]

@andrewfayres, let's tackle this as soon as the gitlab migration is done. I have some rough sekeleton codes for end2end tests