Fixes #12001

Turns out that the order on the protocol for CSS.getMatchedStylesForNode is from least to most specific, so we don't need to calculate specificity ourselves. Only need to filter to the rules we care about (based on what properties they contain) and pick the last.

I don't think adding various tests in the smoke to really confirm this is too useful–we should just rely on it being a given, as DevTools depends on it. To confirm it works, I am first putting up a draft PR that keeps the current implementation and asserts when the simpler approach does not match. This should pass in all the simpler cases we already support, and only fail for the complex stuff we don't yet support. I ran it on cnn.com, theverge.com, and other big sites and got no errors. The smoke tests seo perf also still work.