A Simple and Adaptive Learning Rate for FTRL in Online Learning with Minimax Regret of Θ(T2 / 3) and its Application to Best-of-Both-Worlds