generate_release_data.py: use multiprocessing

Refactor the code: thanks to multiprocessing, extract the languages
for each release using a separate worker.

Depending on the number of (v)CPUs, the speed can increase a lot.
On a test system with 4 CPUs the real execution time dropped
from ~120 minutes to ~40 minutes.