Add San Diego, separate core models from custom models

[pksohn]

Addresses #8. This PR is the result of a pretty thorough audit of this repo, urbansim_defaults, and sandiego_urbansim. I compared all of the model code (orca variables, tables, injectables, columns, etc) to try to arrive at a good "core" set of functions that aren't very idiosyncratic or unique to a single region. Then I separated any functions that were unique to a region (SF or SD) and put them in a custom module for that example region.

For San Diego, I generated a subset of their full dataset, so that it could be included in the repo online (<100MB), but also tested the full dataset with python 2 and 3. Now both sf_example and sd_example directories have runnable demos.

Other things:

• Removed python 3.6 from Travis builds (probably unnecessary at this point)
• Some edits to the readme - I think it's sufficient to just mention something quick about virtualenvs instead of adding to the instructions. I also think dependencies like pip are obvious enough.

I'm happy with this as a starter model at this point, and I think it's a good foundation for further development on the real estate models. This allows us to test with two regions that have slightly different implementations and data.

[pksohn]

Thanks Eddie. I made two changes:

• Move SimulationSummaryData to custom module
• Get rid of unused imports (but keep the "unused" imports that actually perform orca registrations)

I'm leaving the rest as-is because I think they represent good-enough models for typical use, that can be easily overwritten by a custom handler.

[pksohn]

Thanks Martin. Estimation.py and Simulation.py are named that because they were generated from the jupyter notebooks (i.e. Estimation.ipynb) that were already in the San Diego urbansim repo. I left the files the same name just to make it clear they are identical.

I added a line for installation. If this is good I'll merge after tests pass.