On The Expressivity of Objective-Specification Formalisms in Reinforcement Learning