core(trace): compute trace for main frame and any child frames

[adamraine]

Follow up to #11701 and #11713.

@paulirish recognized how loading two pages at once will lead to issues when calculating LCP for all frames #11701 (comment)

This PR makes some substantial changes to the trace processor which allows us to calculate the trace for the main frame and any child frames:

1. Determine the root ancestor frame for each frame using FrameCommittedInBrowser events.
2. Trace events belonging to the main frame tree for the page are put into frameTreeEvents.
3. Use largestContentfulPaint::* events in frameTreeEvents to calculate LCP from all frames. The previous behavior introduced in core(metrics): support LCP in all frames for devtools throttling #11701 looked for NavStartToLargestContentfulPaint::*::AllFrames::UKM events in the entire trace.
4. Use frameTreeEvents instead of the entire trace to calculate CLS from all frames.
5. Update the minified test trace to include FrameCommittedInBrowser events. For future works, this trace can test FCP from all frames, this is so we won't need to update it again when that gets done.

[connorjclark]

LGTM.

An optional refactor idea I leave to you (and @patrickhulce to weigh in) to decide on.

• modify computeValidLCPAllFrames to also return the lcp of each frame
• call computeValidLCPAllFrames just once in computeTraceOfTab
• either pass the main frame result to computeKeyTimingsForFrame; or circumvent that fn entirely and set the appropriate return values directly in the return statement of computeTraceOfTab

[adamraine]

• modify computeValidLCPAllFrames to also return the lcp of each frame
• call computeValidLCPAllFrames just once in computeTraceOfTab
• either pass the main frame result to computeKeyTimingsForFrame; or circumvent that fn entirely and set the appropriate return values directly in the return statement of computeTraceOfTab

I like explicitly computing the main frame LCP separately in computeKeyTimingsForFrame and passing the main frame result as a parameter just to return it seems weird.

I'll try to refractor this in the next PR for FCP since we would probably want a similar change for that event.