Approximate Bayesian Inference for Latent Gaussian Models Using Integrated Nested Laplace Approximations