This function takes numbers that are very small and sets them to the minimal tolerance for C++.
Doing this prevents NaN from entering the optim routine.
Usage
code_zero(theta)
Arguments
theta
A vec that contains estimated GMWM theta values (untransformed).