But IP geolocation has pretty bad resolution. Generally, it can reliably locate country, but rest is a wild guess. E.g. my IP is geolocated to city 300km away (as that is where my ISP has business address).
Yes, because most users will want low latency more than they want a hard-to-geolocate IPv6 address; IOW they will just use the IPv6 addresses their ISP assigns them, which will be easy to geolocate even without the ISP's explicit cooperation. Worse still, because IPv6 promises to eliminate NAT, geolocation will likely become even more precise as user devices will have global addresses that they are not sharing with anyone else.
How can per-device IPv6 addresses alone enable better tracking if they’re long-lived and the device moves physical locations during that time? How would a tracker with only IP address know that the device moved?
A number of ISPs have started using NAT for IPv4 due to the address space crunch, which means that at best those IPv4 addresses can only be used for coarse geolocation. The fact that IPv6 addresses are per-device means that at a minimum you can get household level geolocation for each device (based on the prefix). The fact that addresses are long-lived is not really relevant, since addresses are only long lived with respect to a prefix and prefixes will change when a device moves to a different network (almost always the case with residential service).