add parallelization to centroid_calculator using joblib

Created by: bj-s

using joblib in centroid calculator in a current FLASH beamtime I achieve a centroiding speed improvement of factor 9-10 and consequently overall for the full post-processing a factor of ~5. (using n_jobs=20)

To confirm the code works properly, for a sample run produced file size and computed md5sum was compared between unparallelized and parallelized version.

This parallelization makes use of the already implemented option number_of_processes to control n_jobs option of joblib.Parallel property. For number_of_processes=1 the previous non-parallelized function is applied. Therefore also changed the default value for number_of_processes from -1 to 1 in main.py.

joblib was added as a requirement to the setup.py.