Assessing the validity of NMR relaxation rates obtained from coarse-grained simulations of PEG–water mixtures