tests(smoke): update CLS-AF expectation

[adamraine]

CLS weighted scores were erroneous and should be fixed in ToT:
https://chromium.googlesource.com/chromium/src/+/042fbfb4cc6a675da0dff4bf3fc08622af42422b

[brendankenny]

Interesting that this affects us. From the change description, at least, I wouldn't expect handling zoom would change things for our smoke test.

[adamraine]

This was the full bisect list:
https://chromium.googlesource.com/chromium/src/+log/02c23acd7d6c319ae7976208159e202abd246934..8384f1379a4346fc78da6945a335d947e75632f8

I didn't notice any other suspect changes.

[brendankenny]

oh, I think you're definitely right, I'm just trying to understand why our number would change. I have no idea what the larger implications of PropertyTreeState::Root() vs frame.LocalFrameRoot().ContentLayoutObject()->FirstFragment().LocalBorderBoxProperties() are, though :)

[brendankenny]

There's some weirdness here that @adamraine, @paulirish, and I looked into a little bit but haven't figured out yet. e.g. when I try out the crbug's demo https://output.jsbin.com/wegixex using m91 and m92 (where the change landed), I'm not getting different CLS score or weightedScoreDelta in Lighthouse (have to hack around NO_FCP errors). If this change implies that our defaults run into a CLS change with m92, I would have expected a change on that page as well, but maybe there's something about the iframe relative sizing that just happens to work out?

However, when we've landed clsAF smoke changes before (#11874 and #12034) I don't think we actually verified the numbers, we just trusted what Chrome gave us. Should we do the same here to unblock CI and file a follow-up issue to make sure we do our due diligence before making all-frame stuff visible in the report?

[brendankenny]

when I try out the crbug's demo https://output.jsbin.com/wegixex using m91 and m92 (where the change landed), I'm not getting different CLS score or weightedScoreDelta in Lighthouse

@patrickhulce pointed out that the demo page has a <meta name="viewport" content="width=device-width"> but our smoke test does not, which would affect zooming. Adding a meta viewport to the smoke test makes it so CLS doesn't change from m91 to m92, so that probably solves that mystery.