First paint is not consistent

I believe the first paint benchmark is inaccurate as it fluctuates significantly between runs, locally and in the published results.

Here are screenshots for the same vanilla and svelte implementations in the keyed table between two runs:

Chrome 142

Chrome 144

If this was caused by the difference in Chrome versions you'd expect vanilla-lite to stay where it is.

For an example within the same Chrome version, the current results shows Wallace keyed has a better score:

Than Wallace non-keyed:

Which, because I know exactly what is in those bundles, should not be the case. In fact you can see that non-keyed has a slightly smaller bundle size.

This can be easily reproduced by copying a few implementation directories and running them alongside their originals, which should produce the same results. I know there's going to be some variation between runs, but first-paint seems to vary significantly.