Global Error Bounds and Linear Convergence for Gradient-Based Algorithms for Trend Filtering and $\ell_{1}$-Convex Clustering