Shapely: Automatic doc builds

Incidentally, matplotlib does this already. See https://github.com/matplotlib/matplotlib/blob/master/.travis.yml#L126

I definitely think the process could be improved though...

Additionally, I just found out that ReadTheDocs can now do conda installations, making RTD a viable option for Shapely doc building (I did a proof of concept for Cartopy with some success).

I've had a go at this and have something working, based on this tutorial: https://gist.github.com/domenic/ec8b0fc8ab45f39403dd

The changes are here: https://github.com/Toblerity/Shapely/compare/master...snorfalorpagus:autodoc

The process requires you to create a new SSH key pair and put an encrypted version of the private key into the repo / Travis.

It's set up to only build the documentation in one instance (the Python 3.5 with speedups). https://travis-ci.org/snorfalorpagus/Shapely/jobs/177188968

This has highlighted that quite a few of the scripts that create figures in the documentation aren't compatible with Python 3.x, e.g. buffer.py which gives:

Traceback (most recent call last):
  File "buffer.py", line 29, in <module>
    ax.set_xticks(range(*xrange) + [xrange[-1]])
TypeError: unsupported operand type(s) for +: 'range' and 'list'

@snorfalorpagus before you spend more time on this, let me look into whether our manylinux1 wheels help make readthedocs a better option.

Also want to raise awareness of https://github.com/drdoctr/doctr if you do go down the docs on travis route.

I'm moving this one to the next milestone. I'd like to unblock 1.6.0 and work on the doc infra after.