Levenberg-Marquardt algorithm now available and some minor fixes

[zhucaoxiang]

Levenberg-Marquardt algorithm is now available. The calculation of chi-squared and the packing of jacobian are different from the present optimization algorithms (DF, CG). Please view #21 for more details.

The L-M optimizer is controlled by the following input variables:

LM_maxiter     =        20             ! maximum iterations allowed for using Levenberg-Marquard (LM)
LM_xtol        =        1.000D-08      ! if the relative error between two consecutivec iterates is at most xtol, the optimization terminates
LM_ftol        =        1.000D-08      ! if both the actual and predicted relative reductions in the sum of squares are at most ftol, the optimization terminates
LM_factor      =      100.0            ! the initial step bound, which is set to the product of factor and the euclidean norm of diag*x if nonzero