Merge branch 'fix/parallelism-and-logging-issues'

fix: efficiency improvements

fix: progress logging

fix: potential fix for issues with parallelism

just pushing everything again

moved policy updates to gpu, good speed increase!

lot of experiments, reworked reward

logging of completed tasks and cases

sharing distribution stats as input

train-test split, new reward function using finished cases at step

train-test split, duration as NN input instead of timestep, cleanup

cleanup and implement learning logic

remove blank lines

working distribution sampling with min/max values and a bit of cleanup

add capability distribution