Run benchmarks on CI

[iamed2]

We should be able to run benchmarks on CI infrastructure and report on regressions or progressions. We may need to widen tolerance to account for variance in the cloud, but Curtis Vogt (@omus) has determined that it's feasible to rely on Travis-run benchmarks most of the time.

[omus]

Just going to note that benchmarks on Travis CI do give a "ballpark" view for performance regressions. Definitely not great at finding subtle regressions.

[iamed2]

Mhm, they'll need to be run more seriously in a controlled (local) environment when examining.