Improve reporting and usability of UI benchmarking

[chikeichan]

We record various page load metrics on each Pull Request, added in #7871

However, as pointed out in the comments by @Gudahtt, these measurements are not stable and often vary wildly from build to build. We also do not record this data over time to get an eye for the impact of changes.

• Investigate and stabilize large variance in load times (make them more reliable)
• Track metrics overtime so that large changes in load time can be flagged
• Automate benchmarking & integrate with build

[brad-decker]

@danjm @Gudahtt do we think this is more of a tooling/testing (like using lighthouse or something in the build process) or an analytics/metrics issue?

[Gudahtt]

The former. And actually we have already implemented this: #7871

Though in practice it's not very useful yet because we don't keep track of this over time, which we'd need to start doing in order to spot regressions. There are also a lot of improvements we could make to make the measurement more stable (it can vary based on various factors at the moment that are outside of our control, so we might get "false positive" performance regressions).