Speed up SICOR for EnMAP

There seems to be a lot of room for speeding up the entire processing in SICOR for EnMAP.

• make_ac_enmap() takes around 5 min per detector but runs in singleprocessing
• the water vapour first guess retrieval takes around 2 min
  • multiprocessing is disabled for Windows (#42)
  • Mac OS uses 'spawn' as default multiprocessing backend which might be slow with globals.
• consider switching from multiprocessing to joblib which would make the implementation much easier
• there are a lot of nested for-loops in the code, many of them can for sure be optimized or vectorized