I think the definition of "expert" can be pretty ambiguous. We have one type of "expert", the MLB hitter who is an expert because he is more skilled than others at hitting. And we have another type of "expert" who is an expert because she knows more about something than others. These are two very different types of "experts" but what makes them experts is their skill/knowledge relative to the general populance rather than skill/knowledge relative to some absolute.
That's true. I don't see any reason why they couldn't cache non-amp content with a combination of checking for speed benchmarks and schema markup. When you think about it that way, it seems like they are more concerned with controlling the user experience than they are with speed improvements.
Google already has your IP; it's their page. Preloading resources from it's own CDN doesn't tell them anything they don't already know. Preloading resources from someone else's domain would.
Google can't "simply preload resources of non-AMP pages". Using a CDN means they can get more consistent load times and avoid the inevitable privacy concerns that come with loading 3rd party content the user may never actually click on.