So, let's take the situation where a perfect cylinder or ball is rolling, without slipping, to the right. The ball is rolling clockwise from our perspective viewing it rolling to our right.
There's a static friction force at the point of contact, preventing relative motion of the instantaneous point of contact.
From the center of mass perspective, this friction force is the only horizontal force (and the only net force) on the rolling object. So, it slows down and stops.
From the torque about the center of mass perspective, this friction force exerts a torque about the C.O.M. The direction of this torque would increase the angular velocity in the counter-clockwise direction — so the ball would increase in linear speed if this were true.
I suspect the paradox is trying to evaluate the torque about the C.O.M when the fixed point of rotation is elsewhere? Is there a pseudo-torque? Do truly ideal balls never stop rolling?
The math works itself out from either perspective if there's an incline...