Make research_VENUS.load_VENUS() fault tolerant

[douglatornell]

Original report by Doug Latornell (Bitbucket: douglatornell, GitHub: douglatornell).

---

load_VENUS() uses request.get() without any re-try log. It almost always works (and is fine for notebook use), but when a ConnectionError occurs in the nowcast system's use case it stops the automation path to pushing the results to the web (happene on 11dec15).

Idea:

• Split load_VENUS() to separate getting the data from the ONC web service and parsing it into a pandas dataframe.
• Retain a load_VENUS() function that gets data from the web for use in notebooks.
• Change compare_VENUS() to accept either a buffer containing the web data, or a pandas dataframe as an arg.
• Do the data download using nowcast.lib.get_web_data() in make_plots._make_research_plots() and pass the data into compare_VENUS().

Need to think about/explore binary vs. text data handling as nowcast.lib.get_web_data() presently only handles binary (response.content).