Parallelizing the "Collection processing" code

[NumerousHats]

Thank you for a fantastic resource! These Jupyter notebooks have been extremely useful in getting me comfortable with using librosa and python for audio. Using frequency rolloff and contrast stats at collection-level is a particularly useful idea. Since, in my hands at least, the librosa methods are CPU-bound, the processing of large file collections can be greatly accelerated by using multiple cores in parallel.

I have modified the code to use multiprocessing: https://gist.github.com/NumerousHats/e701db2e04b81b7718bf247d78560df6