Differentially Private Gradient Flow based on the Sliced Wasserstein Distance for Non-Parametric Generative Modeling