Long-run Behaviour of Multi-fidelity Bayesian Optimisation