moving training into it's own method

[durandom]

The idea is to trigger the app via an env variable to perform training and store the output on storage

[durandom]

@TreeinRandomForest merge if you like this.

I would like to keep the clustering/train and clustering/infer parts to you and I focus on the plumbing My next step would be to make the training possible on ceph and S3 and deployable on openshift. But on later PRs