Many Optimizers But Only One Training Path: Repeated Resampling for Adaptive Optimizer Selection