Fixing Travis builds

[smmaurer]

This PR fixes the Travis builds.

Description

It seems like a problem emerged in the last few months, related to including pip in the conda environment. Example of a failed build: https://travis-ci.org/UDST/pandana/builds/557649396

There's actually no need to install pip like that; it's there automatically. This PR removes it from the script, and the builds work again.

I also switched the Python 3 environment from 3.5 to 3.6, which is the current stable release.

Should we do some additional Travis cleanup?

I would be in favor of overhauling the Travis script so that it builds Pandana from source the same way users will -- something like pip install ., without using conda at all.

This seems like a better test of the build process and requirements list than the current script, and it would run faster too. Should i try this out?

Here's an example of a much cleaner Travis script: https://github.com/UDST/choicemodels/blob/master/.travis.yml

[coveralls]

Coverage remained the same at 91.089% when pulling c073d5da3b35be3d9b385c2ac5df9945347fc3c2 on travis-fix into 961a7ef8d3b0144b190cb60bbd61845fca6fb314 on develop.

[coveralls]

Coverage remained the same at 91.089% when pulling c073d5da3b35be3d9b385c2ac5df9945347fc3c2 on travis-fix into 961a7ef8d3b0144b190cb60bbd61845fca6fb314 on develop.