No paradox here? Improving theory and testing of the nuclear stability–instability paradox with synthetic counterfactuals