The goal of Cloudflare’s fingerprinting is to detect whether a user agent appears to be a legitimate human. It’s not to identify human users across websites.
Every HN thread is full of people who think webmasters should just pay through the nose to handle bot traffic to preserve the sacred rights of turbonerds to visit their website using Lynx on their toaster.
I should think that there should be a better way (e.g. port knocking, instructions for manually correcting the URL that cannot easily be automated, additionally supporting alternative protocols, etc).
Why not? PoW challenge doesn't whitelist botnets. If the dumb scraper makes only get requests and doesn't solve the challenge, it doesn't matter how it connects, even if it's a perfectly hidden tor exit node.
Because the work would be done by the compromised residential device. No bothnet owner is going to care if their 100,000 rooted routers have to do a little more work. It’s still “free” from their perspective.