Model-Parallel Fourier Neural Operators as Learned Surrogates for Large-Scale Parametric PDEs