Mata implementation of Gauss-Legendre quadrature in the M-estimation context: Correcting for sample-selection bias in a generic nonlinear setting