Marginalising over Stationary Kernels with Bayesian Quadrature