Adaptive Selection of the Gaussian Sampling Scale for Monte Carlo Estimation of Loss-Landscape Convergence