I think there are two primary reasons.
1.) A developer wants to develop/test against an x.y release that only changes minimally (major bug and security fixes) for an extended period of time.
2.) The point release model where you can decide when/if to install upgrades is just "how we've always done things" and a lot of people just aren't comfortable with changing that (even if they effectively already have with any software in public clouds or SaaS).
I largely agree with your other points.
Re: point 1, I'm definitely aware of that need but the only cases I see it are commercial settings where people have contractual obligations for either software they're shipping or for supported software they've licensed. In those cases, I question whether saving the equivalent of one billable hour per year is worth not being able to say “We test on exactly the same OS it runs on”.
I'm not sure there are enough people left who have the “don't touch it for a decade” mindset, aren't working in a business environment where they're buying RHEL/SuSE/Amazon Linux/etc. anyway, and are actually going to contribute to the community. 100% of the people I know who used it were doing so because they needed to support RHEL systems but wanted to avoid paying for licenses on every server and they weren't exactly jumping to help the upstream.
Red Hat bought CentOS in the first place because they were having trouble attracting volunteer labor and I think that any successor needs to have a good story for why the same dynamic won't repeat a second time.