Long-Run Behaviour Of Multi-Fidelity Bayesian Optimisation