Is there a way to set upper/lower bounds or general linear constraints to the coefficients of scipy's/Numeric's linear least-squares minimization routines (e.g., lstsq)?