So people can pretend the "abusive" user (e.g. ones posting "harmful emojis") doesn't exist.
It's a mindset, I think. See, for instance, this woman complaining about News Genius[1], discussed here [2]. At one point, she's upset that a blocked Twitter user can see her public tweets.
People should be able to moderate and exclude people. It's just funny how powerful they think that capability should be.
Probably because the only technical advantage to the "fork" button is that it tells all sorts of GitHub social-coding backend things (pull requests, the network graph, etc.) that the two repositories are related, and it is exactly those things that you'd want to block. If I were in their shoes as an engineer, I'd probably decide that it's more reliable to just disable the fork feature entirely, than to go through everything that fork does and disable it. (I'd consider implementing a second "fork" codepath that just does the equivalent of git clone && git push, and silently replacing the regular fork button with the fake-fork button instead of disabling it, but chances are I'd probably find this lower-priority than other work.)
Blocks from users to other users (which have been around just about forever) work the same way.