Optimization Landscape of Policy Gradient Methods for Discrete-time Static Output Feedback