Accelerating Physics-Informed Gaussian Processes: Enhancing convergence and efficiency in Bayesian partial differential equation learning