Identifying the origin of differences between 3D numerical simulations of ground motion in sedimentary basins: lessons from stringent canonical test models in the E2VP framework