Automate the evaluation of matching and searching algorithms

[platers]

In order to improve search performance we need an automated evaluation suite. I made a first attempt at this in eval.py, where I can manually add hard test cases to debug.

A more systematic, larger, test set is needed to count false negatives and positives.
Maybe can be obtained from the musicbrainz database.