I get it that CSAM within the Apple ecosystem is probably a problem, possibly even a problem aggravated by Apple's heretofore good privacy practices. That said, on-device surveillance, even surveillance with as many privacy protections as Apple tried to build into their recently-announced-but-now-delayed system, is a non-starter for me... if they eventually go forward with this I'm out, full stop. I'll switch to LineageOS or a PinePhone, and fall back to FreeBSD. I have firsthand experience with US intelligence hacking a computer I was using because of my political views in the early 2000's; it was crude back then, but it happened.
I think that Apple need to go back and re-visit what is the problem they're trying to solve here? Since they have said that the scanning/hashing occurs prior to upload to iCloud it seems reasonable to assume that their goal is to disrupt sharing of CSAM. It seems to me that would most likely be happening via Shared Albums in Photos; I would be fine with server-side scanning of Shared Albums, since that is what every other cloud vendor is already doing. I really do think that in this deeply polarized modern environment (red vs blue, vaxxed vs. unvaxxed, etc) the slippery slope argument really does apply - at some point on-device scanning for CSAM will turn into on-device scanning for evidence of wrongthink.