Quantitatively predicting the adsorption isotherm of water in H-ZSM-5 using reactive machine learning potentials