Benchmarks can silently fail

In the original Convex Hull benchmarks that I added (#51 and fixed in #55), I made a mistake where my number of points was a nice power of 10, which is unfortunately not a multiple of 3, and produced an error that later would stop the run of the intended benchmark.

Confusingly, the numbers still made sense somewhat, so it wasn't immediately obvious that the benchmark was bogus.

Also, since the benchmark "game" window starts fullscreen, it covers the usual debugger tab that would be red and maybe get your attention.

Mistakes like the described above might possibly be catched when adding the benchmark / code review. But it would be nice to detect script errors when running the benchmark: maybe a future (unreleased hopefully) version of the engine mistakenly breaks some feature, and currently the only way to spot it looking at the results would be if the time noticeably changes (likely).

---

So, what would be a good way of detecting script errors when running a benchmark?

We can add a boolean like `ran_ok` in the JSON, and another column with "OK" or "ERR" on the GUI.

Start by tracing how benchmark script errors affect the generated JSON and how results are displayed in the GUI. Reproduce the malformed benchmark case described in the issue, then define and implement a reliable error indicator. Done means failed benchmarks are distinguishable from valid results without relying on timing changes.

Written by the indexing model from the issue text.