A few examples that come to my mind: Current battery level, typing speed, phone model, OS version, battery wear (guessable from charge/discharge rate), charge rate (depends on the charger), finger width, propensy to use features differently (scrolling, zooming, selecting text), clipboard contents, RTC lag, microbenchmark performance... System preferences alone is probably enough (brightness level, airplane mode, dark mode).

If a fingerprinting library is present in more than one app, I find it unlikely that both fingerprints couldn't be linked to the same user. If one of these apps has a log-in, they can probably link that up with the rest of browsing history, on-device and cross-device.

Sounds far-fetched? I don't think so. I rather think I'm underestimating the issue.

I think one technical answer to that would be to "taint" every measurement of a potential identifier, and track its usage across the program. If the app tries to submit information that is somewhat related, block it.

Alternatively, compute a score and block it above a certain threshold (that will be gamed, but could help a transition). Or use a RR-like mechanism to change the measurement to a dummy value, and replay up to the exfiltration point.