Molecular-level operational optimisation of hydrocracking process via kinetic-informed deep-learning surrogate modelling