Strategies for Safe Multi-Armed Bandits with Logarithmic Regret and Risk