Dynamic Regret of Adaptive Gradient Methods for Strongly Convex Problems