On the duality between locally optimal tests and optimal experimental designs