And actually, when I started programming professionally, I didn't think that management was necessary, and that we could self-organize and all that. But as I changed projects and bosses, it turned out, I just had a good manager (and it was a waterfall-like method, by the way).
So I think the idea of self-organization is intuitively appealing to many people (especially young without any experience), and if you have a person in the team who can naturally take each of these roles, then I think it can work (that is - sometimes). But today, I don't think it will work in general, and for that reason.
Now I recognize that, for instance, you need somebody well-organized on the team who can make sure things do not fall under the table. Or you need somebody who is process-oriented, interested in improving process. Then self-organization can work.
If you don't have people with specific traits like that, or they don't emerge as natural leaders (gain respect) in their specific areas of interest, then you need to appoint somebody to have that role. Then it makes sense to have explicit project manager, or product manager, or architect, who focus on certain tasks and concerns.
I suspect the fact that sometimes you can do without having these people designated as such can account for some anecdotal success of Scrum (or any other method - I personally had the best experience with waterfall-like approach).
Not really, and I don't think the industry as a whole does either. Everyone has their own set of tasks and concerns; I don't think there's any consensus that says that all the important ones are on your list, or that everything on your list is important.
That's one way to break down the set of tasks/concerns you've identified, sure - but only one. I don't think that the industry as a whole accepts your categorisation into three, or even your paradigm for which tasks do and don't need addressing in this way.
You're coming across as far more ideological than the thing you're attacking, assuming that your particular breakdown must be the canonical one when it simply doesn't have a level of industry acceptance that would justify that.