WIP: Skeleton workflow

[kmuehlbauer]

This is a attempt to setup a skeleton notebook workflow.

The suggested workflow (heavily borrowed from wradlib) executes the notebooks in a headless state using Travis-CI and creates a sphinx-doc from them.
The notebooks need to be in a clean state without output. I manually pushed the created docs to the gh-pages branch, they are available here: http://openradarscience.org/openradar-cookbooks. If merged, the docs will be build automatically and pushed to the gh-pages branch.

The notebooks will be placed under the notebook-folder inside the repo. We can have different subfolders, corresponding to different aspects (wradlib, pyart, erad, ams etc.). One thing we need to do manually is reflecting the notebooks structure inside the doc/source as seen with the skeleton. Incorporation of the data will be done with the openradar-data repo (not yet functional).

Travis-CI is momentarily using latest miniconda but we can think about other means (vagrant/docker) to get all dependencies in one (eg. for baltrad).
The workflow will then be:

1. Fork the repo(s)
2. Add data -> PR -> MERGE
3. Add notebooks
4. PR - The notebooks will be executed, if anything brakes we get informed by Travis-Ci.
5. Review -> MERGE
6. Executing notebooks, docs-building and uploading
7. Commit is reflected in the docs index.html

Within Travis-CI python 2.7, 3.5 and 3.6 are tested. The docs are build atm on 2.7.

[kmuehlbauer]

@openradar/developers With ERAD2018 approaching with lightning speed, I'm would here your opinions on this. It would be a good opportunity to fill this notebook with something special to advertise at ERAD.