A new use of Douglas–Rachford splitting for identifying infeasible, unbounded, and pathological conic programs