Create doc page to collect benchmarks using dirtyduck data, generate initial benchmarks

@nanounanue is already working on this.

Since Mar 1, 2018.

Using the open dataset and experiment config in https://github.com/dssg/dirtyduck we should start to collect benchmarks when new features are introduced. Although collecting such benchmarks is an ongoing process that cant be covered by one unit of work, this issue tracks the creation of such a page to collect them with some examples. Some ideas for different benchmarks we could produce on master: s3 vs FS storage, single threaded vs n processes in a multicore experiment, small/medium/large grids.

Given that memory is often an issue, it could make sense to try and measure max memory usage as part of this benchmark, not just runtime. I'm not entirely sure how to do this, but if nothing else the memory/swap available on the test instance should be noted (along with other relevant instane info, of course)