Atomistic simulations of plasticity in topologically close-packed phases using machine learning potentials