However my argument is specfically to the OPs point that they should just scan the photos "in the cloud" and not "on your device". But the reality is they were only scanning photos that were being uploaded to the cloud, they were just doing it on your device so that Apple doesn't need to have a server that can decrypt your photos to do it in the cloud. Which is what most others (including Google) do.
I think you might have misinterpreted whatever those analysts wrote. The chances of thirty images all coincidentally matching both perceptual hashes is essentially nil.
The human reviewer would've gotten "reduced quality images" of the matches and would make a decision based on those.
And yes, people could've just turned off iCloud. That's how encryption works.
What are the benefits? It would tell Apple (and hence law enforcement) who had copies of existing, known CSAM. I guess that would help catch such people. But such individuals would also have strong incentive to simply turn of iCloud to completely avoid detection.
What are the risks? Apple employees rummaging through your photos (after 30 NeuraHash matches, a human reviewer would go in and check your photos, and as some analysts pointed out, hash matches can be spurious, that is, completely different photos can have the same hash by coincidence).
Over time, would the program expand beyond CSAM? Consider the intersection of countries in which Apple sells iPhones and countries that criminalise being LGBTQ [1]
[1] https://www.humandignitytrust.org/lgbt-the-law/map-of-crimin...