Perhaps its a bad article, but I can't see the value of solving the boson sampling problem by sampling bosons having any application in general computing.

I think "propaganda" is too harsh, more like "spin". This is apparently a fine achievement of quantum optics that will hopefully be applicable to quantum computing. But it is being directly billed as a quantum computer itself, which is a stretch.

A classical computer is more than an experiment. The "just" part of that statement is what matters. Computer implies some degree of generalizability in computing things. If it can only compute one thing, it's at the trivial extreme and would be more logically described in terms of that single things it does. For example a beamsplitter that divides power in half. We call it a beamsplitter, not a classical computer that calculates 1/2 input power.

What's the advantage of the abacus or the Enigma machine? Or any of the many Turing-incomplete machines that were built before fully general computers in the 1940s-50s.

This talk might enlighten you https://www.youtube.com/watch?v=njwQgz63rIs

A computation model doesn't have to be Turing complete to be useful in practice. Consider a language where all programs must halt - all loops must be bounded, and recursion is not allowed - you can still solve a lot of problems with a language like this, but you cannot say, run Brainfuck, as such a language is not Turing complete.