Why Most Optimism Bandit Algorithms Have the Same Regret Analysis: A Simple Unifying Theorem