Improved Regret for Zeroth-Order Adversarial Bandit Convex Optimisation