Set up minimal reproducible component

[richfitz]

Obviously full analysis is far too much for Travis, but might be worth setting up a minimal set, following along from some of the wood ideas.

• Set up environment
• Download and process data (use cached versions from gh)
• Run ML on one tree with adequacy on (say) 100 points, same for MCMC.

[richfitz]

Hmm. Running this at the moment - this might be too slow for practical use. Just sorting the data out takes about 10 minutes. If I had more time I'd look at speeding up the bottlenecks there.