A Layer-3 (network-layer) blacklist risks cutting off innocent CGNAT and cloud users. What you’re proposing is similar to mechanisms that already exist (e.g., access control lists at the ISP level work by asking computer B which requests it wants to reject and rejecting those that originate from computer A). However, implementing any large-scale blocking effort beyond the endpoint (i.e. telling an unrelated computer C to blackhole all requests from computer A to computer B) would be too computationally expensive for a use case as wide and as precise as “every computer on the Internet”.
Also, in your post you mentioned, “A host would need to have a way to identify itself as authoritative, responsible for the IP address in question.” This already happens in the form of BGP though it doesn’t provide cryptographic proof of ownership unless additional mechanisms are in use (RPKI/ROA).
See https://coveryourtracks.eff.org/ with Tor set to the Safest setting. The user share for Tor might be very small. However, because all Tor users have the same configuration, it doesn’t matter whether a fingerprint differs from Chrome. Among the x% of Tor traffic, x% traffic shares the same fingerprint. Chrome might account for y% of the traffic where each user has a unique fingerprint. But as long as x is not negligible, the fact that you’re using Tor provides very few bits of information (as an example, about 8 bits of identifying information) compared to a unique fingerprint (which provides much more information). I agree that Tor is not without its flaws, but saying that Tor deanonymizes you because of its user share is wrong. Also, please note that the EFF link I shared may be biased in the data it collects.