Bootstrap Confidence Intervals In Nonlinear Regression Models When The Number of Observations is Fixed and The Variance Tends To 0. Application To Biadditive Models