added formatting to verbose and logfile, fixed multiprocessing porblem and GPU usage in analysis_example

[gwirn]

• README
  added conda install instruction to manual installation guide in the README
• environment.yml
  added openmmtorchplugin
• analysis_example
  added main function and if __name__ == "__main__" to avoid multiprocessing error, and added sending the net to the GPU as well as always choosing GPU if one is available instead of manually
• trainer
  prettier formatting in the verbose training setting like

train_mse_loss     : 0.003045
train_physics_loss : 21369.562486
train_loss         : 0.003349
valid_mse_loss     : 0.003196
valid_physics_loss : 22197.075703
valid_loss         : -4.763431
train_seconds      : 9.927747
valid_seconds      : 0.528163
checkpoint_seconds : 0.256874
total_seconds      : 10.712785

and formatting of the log file so it can easily be read with eg pandas.read_csv

[degiacom]

Logging looks nicer like this, both on terminal (in verbose mode) and in the logfile. Having outputs being more human (and pandas) readable is a good idea. The only minor drawback is that this is a breaking update for any previous postprocessing code designed to parse the logfile (this is an easy fix though)! Overall, the benefit of this updated outweighs the small hassle caused by such small fix.