Training neural operators to preserve invariant measures of chaotic attractors