config reshape suggestion

[victor-moreno]

Hi,
A suggestion:
It is a pain to have to either edit the config file or deal with multiple config files because some steps overwrite files of previous steps. This is mainly for crossval/deploy and statistics for crossval, train & deploy. I think the procedure would benefit a lot if a complete analysis can be run with one unique config file from preprocess, if needed, to deployment.

The conflicts mainly arise at deploy. The procedure would benefit for a new config section "deployment:" to avoid conflicts with "modeling:" that contains all the required paths for the deployment step.
This should be very easy to implement, only changing cli.py instructing to read c from cfg.deployment.

statistics should be renamed or changed to a flag (True/False) so that after crossval, train or deploy, statistics are calculated (or not) and saved in an output_dir that do not overwrite each other.