Change our perf comoparison autoloop program to run the benchmarks and generate the real data that the pages build will pick up after each successful run, so that when we merge to main, we will get real data on the pages site.

[mrjf]

On our pages site, benchmarks shows empty because benchmarks/results.json on main is just a stub:

{ "benchmarks": [], "timestamp": null }

What's happening:

1. .github/workflows/pages.yml copies benchmarks/results.json into ./playground/benchmarks/results.json and benchmarks.html reads from there.
2. The file exists on main but with an empty benchmarks: [] array, so the page renders "No benchmark data available yet."
3. You have lots of benchmark programs (benchmarks/tsb/bench_.ts + benchmarks/pandas/bench_.py) — autoloop keeps adding more — but nobody is executing them.
4. benchmarks/run_benchmarks.sh is the script that would populate results.json, but no workflow runs it. It's only invoked manually.

Fix:
Change our perf comoparison autoloop program to run the benchmarks and generate the real data that the pages build will pick up after each successful run, so that when we merge to main, we will get real data on the pages site.