Add proximal adagrad optimizer

[sidgoyal78]

This closes #4687 by adding the implementation of the proximal adagrad optimizer.

The main idea is to modify the proximal gradient descent with adagrad learning rate scheme:

moment = moment + grad * grad
prox_param = param - learning_rate * grad * (1 / sqrt(moment))
param = sign(prox_param) / (1 + learning_rate * l2) * max { |prox_param| - learning_rate * l1 , 0 }

The paper that proposed Proximal GD: http://papers.nips.cc/paper/3793-efficient-learning-using-forward-backward-splitting.pdf
The paper with details of adagrad: http://www.jmlr.org/papers/volume12/duchi11a/duchi11a.pdf

[qingqing01]

Excellent!