Now, it's entirely up to you whether you write that 4-fermion term coefficient as "Riemann tensor (including torsion)" or as "Riemann tensor (torsion free) + (lots of weird, arbitrary-looking interaction terms involving derivatives of B)". So on some level, there's no reason that you must use a connection with non-vanishing torsion. But I would claim that the equations are much more elegant (and give deeper insight) when expressed with torsion "built-in". [Fun fact: So does Polchinski, but he's not talkative about it. If you look up "torsion" in the index of Vol. 2, the first reference is to the page with the equations I've referenced above... but the word "torsion" doesn't appear anywhere in the text of the page!]
Aside: If anyone out there is interested in how this torsion stuff fits into the mathematics of Relativity, you might have a look at my notes on how it would be incorporated into Bob Wald's textbook: http://www.slimy.com/~steuard/teaching/tutorials/GRtorsion.p...
I'm not sure I understand. In Riemannian geometry it might be convenient to use a connection with non-vanishing torsion, but it is not required. What do you mean by 'includes'?