Add travis CI to seaborn

[mwaskom]

This now runs and claims the build is passing, but that's a bit of a lie at the moment. Two reasons:

• ipynbdoctest exits with 0 even when cells do not replicate, so currently things that should count as failures are masked (I'm not sure what happens when cells error)

• one cell per notebook is currently not replicating, with the following message. I cannot reproduce this on my laptop (all cells reproduce). The failure seems to happen near, but not at, the beginning of each notebook. Maybe the first plotting cell?

  mismatch output_type:
  stream
    !=  
  display_data
  

Nevertheless, merging now as it's running and at least the nosetests suite should trigger failures.