Nearly Optimal Algorithms for Private Online Learning in Full information and Bandit Settings