This PR revamps the evaluation infrastructure.

Instead of calling the individual modules (e.g. python -m madmom.evaluation.beats), all evaluation stuff is bundled in the bin/evaluation script. Different evaluation methods are added as sub-parsers, so calling is done via evaluate beats --options path/to/files.

The Sum/MeanEvaluations are also re-organiised to take a list of evaluation objects as their single required argument. They then sum or average the individual evaluations.

As a result, the printing/formatting stuff can be simplified to use simple functions and do not require any formatting classes or the like.