Run benchmarks in CI

[jakebailey]

This adds a simple workflow which benchmarks a change and its parent, displaying / uploading the benchstat results. These benchmarks run on a larger runner.

We'll see how noisy this is given it's on a VM (and we aren't using dedicated hardware for this), but Go benchmarks are a lot less noisy especially given how many runs we can get. (Compared to fully-fledged benchmarks of tsc in JS).

[jakebailey]

The results here show that the workers are not isolated enough from VM noise to be usable, though I will have to try again using dedicated VMs, or see if we can just make VMs large enough to not actually be used by multiple people.