This means, tracking libs can now also access everything of the website they are hosted in, attach any event handlers etc. It is essentially the same as hosting the tracking library from the content server, except that the tracking provider can change the source code anytime without requiring any consent.
This is a security nightmare imo.
At least that's the vision, to make the web "smaller" again like it was before Eternal September, and to create some protocol implementing a vision like that to take over and wrest things from giant oligarchs controlling all social media. I have no idea if making that is even feasible, though I've seen some attempts vaguely similar to that. I see tons of practical challenges with actually implementing such a thing, though, but the idea would be to make all the communication opaque and explicitly take it away from harvesters.
Granted, there's no way that could end things... I'm sure clients might get subverted via various methods to re-enable harvesting but at least it would limit their scope and you could remove such folks as connections if you knew.
I'm surprised they even setup another domain, you could do path rewriting at load balancer level to route the /extras path to the ad tech software / provider.
I think the main point is just to make it harder to block using pi-hole like solutions.