but that instance owners have even more, probably more valuable info, like IP addresses from which not just geolocation but also wake times, device usage patterns and other gnarly stuff could be extracted, that could - together with other personalized surveillance info (like the usual adware stuff) - be aggregated to give a bigger picture.

I have IP behind the geolocation. How do you think that I know the geolocation? It's an IP lookup. My interface that I shown in the image just doesn't publish it because I don't care personally. What I use that service for is simply to track where sensitive emails/documents go. Not to track lemmy. I don't need specific resolutions. Just to know if they leak outside of what I expected.

Device patterns? The app you use is the app you use. That would be given away via your browser header. I also collect that with the tracking image. Just once again. Not shown in the graph cause I don't care to track it personally (I'm only doing this as an example, not to actually aggregate data).

If you use lemmy over the web browser, browsers don't really give up that much information unless you're google themselves. In which case apparently chrome gives up a boatload of information to google's domains.

not-so-public information

You'd have to give me an example of any of what you're referencing. I can collect IP, web headers, access times, and if I tag enough pages or mark the image as non-cacheable could even see multiple views/accesses (you see views higher than actual visitors) I can track your movement across all of the fediverse.

that one can get some info about me through my (public) actions

Simply "viewing" the page (which pulls the image and is not necessarily "public") is a direct rebuttal to obtaining data that isn't "public".