43 multiprocessing training

[jaymody]

Added multiprocessing support for location training. The default number of processors is multiprocessing.cpu_count() - 2.

This PR also adds a argparse CLI for predictions.py to make it easier to run with different options and configurations. Here are some examples: Download new data and train new locations models

python predictions.py --download_new_file true --locations true --train_new_model true 

Train model with train_windows and number of forecast days as 6

python --train_window 6 --num_forecast 6