reader: event identifers building with multithreading

[mikael10j]

During the building of event identifiers we may apply some intense filters. It would be nice to do it in a multithreaded way to save time.

[TjarkMiener]

Hi @mikael10j
Thanks a lot for the issue and the suggested solution. Some points: With v0.10.0, the DL1b data level is included in both the official CTA DL1 data model and the deprecated DL1DH data format (supports multiple data of the DL1b level). The building of the example identifiers is rather quick and can be even bypassed by saving the example identifiers to disk when running more than one random seed. Besides this, Johan had some bad experiences, when working with the deprecated DL1DHWriter (using 'parallel' mode via Python multithreading) on the grid back in the days #51. Since most of our future experiments will be run on GPU clusters, we would probably face similar problems, even though one can also select more supporting CPUs.