Breaking GEO Impasse Amid AI Large Model Iteration: Reconstructing Optimization Logic with a Dynamic Adaptation System