In a CSP encoding of a planning problem, why does an effect constraint involve both state variables at the current time step and the state variables of the next time step? (That is, why is it not enough simply to constrain the values of the variables at the next time step whose values are changed by the give action?)