Global Inducing Point Posterior Approximations for Federated Bayesian Neural Networks