Breaking scaling relations with inverse catalysts: a machine learning exploration of trends in $\mathrm{CO_2}$ hydrogenation energy barriers