Biodegradable magnesium (Mg) implants combine bone-like mechanics with in vivo resorption, yet predictive design requires robust computational links between corrosion-driven degradation and healing. We present a multiscale framework that couples implant dissolution to mechano-biological bone remodeling and, in line with local--nonlocal integration goals, uses two interchangeable corrosion descriptions: a local KKS phase-field variant and a peridynamic formulation. Their consistency is established via a convergent nonlocal extension of KKS that is well-posed, recovers the classical limit as the interaction horizon vanishes, and is compatible with peridynamic kinematics; pitting-style benchmarks verify $L^2$ convergence of the nonlocal discretization to the local counterpart while capturing regularized damage evolution. The corrosion solver is coupled to healing through time-resolved $\mathrm{Mg}^{2+}$ release that modulates osteoblast differentiation and mineralization, enabling joint prediction of Mg dissolution, volume loss, and remodeling-driven stiffness recovery. For efficient parameter studies and longitudinal forecasts, we augment the workflow with a learned time-conditioned diffusion surrogate with super-resolution refinement and uncertainty estimates, yielding a scalable pipeline for multiphysics simulation and design exploration of bioresorbable Mg implants toward patient-adapted treatment planning.