Near-optimal Thompson sampling-based algorithms for differentially private stochastic bandits