On the Best Euclidean Fit to a Distance Matrix