Add benchmark models

[junpenglao]

Thanks to the work from @ColCarroll and @TomAugspurger in #1829 we now have a way to benchmark: http://pandas.pydata.org/speed/pymc3/

We should port some of the models from our docs to have more benchmark model to better cover our code base. The following comes to mind:

• Variational Inference
• Mixture model
• GP
• Same model with different sampling initialization (e.g., adapt_diag vs adapt_diag_grad)

[aseyboldt]

It would also be great if we could report effective samples per second for some models. Or is the benchmark limited to looking at the time?

[ColCarroll]

That's a great idea. I don't think you could benchmark both time and effective samples in one go, but looks like it would be added under the "generic tracking":

http://asv.readthedocs.io/en/latest/writing_benchmarks.html#tracking-generic

I may add another instance of the radon model to watch ESS.

[springcoil]

I made a first stab at the mixture model. I hope that'll work.