Add new node implementing Jaakkola & Jordan’s lower bound on sigmoid (log-sigmoid) in Bishop PRML

[mhidalgoaraya]

Introduce a new node into the codebase that implements the lower bound on the log-sigmoid (sigmoid) function as described in Pattern Recognition and Machine Learning by Christopher M. Bishop, following Jaakkola & Jordan (2000).[Page 497 equation (10.144)]

Specifically, this lower bound has a functional form using a Gaussian approximation, obtained by making transformations both of the input variable and of the sigmoid function (per Jaakkola & Jordan).