performance-impl: use fcp and ofv entryTime when summing for cls-fcp and cls-ofv

[samouri]

summary

• All CLS-related events are emitted within the same flush(), at the time when CLS would normally be emitted.
• cls-fcp and cls-ofv now use the corresponding entryTime for knowing which layout-shifts to include, as opposed to assuming they were invoked synchronously after the corresponding event.
• In order to accomplish both items above, we now collect all the layout shift events into an array instead of summing as we go. I've added a limit of 1000 entries so that it can't have unbounded growth.

[samouri]

Note: most of the random-seeming number changes in the tests are reverting the previous PRs changes

[kristoferbaxter]

Looks good. Thanks for addressing quickly.