Run benchmarks on push to default branch

[LilithHafner]

It is sometimes useful to run benchmarks on every commit to the default branch (posting comments on commits) in addition to on pull requests. The results may differ from those posted to the pull requests if the pull request head is out of date.

[MilesCranmer]

One tricky part about this is that the variability in performance across runners tends to be larger than variability over a single action. The action runner is some small virtual machine and while even a single action is pretty noisy, multiple actions will be noisier since it might be a completely different machine. But worth thinking about, and maybe it's not as big a deal as I think.

[LilithHafner]

Good to know. Whether that noise is an issue likely depends on the benchmarks being tracked.

FYI I've implemented this here: LilithHafner/DynamicDiscreteSamplers.jl#50 (Example comment: LilithHafner/DynamicDiscreteSamplers.jl@79e5728#commitcomment-151878662). We'll see if it's actually useful.

[LilithHafner]

It is very low-noise which is nice.