Add ADAM optimization in STM for Gaussian Process training

[somu15]

Reason

Currently Gaussian Process (GP) training utilizes (mostly) CG optimization through TAO. Adding ADAM optimization since its popular for GP training and permits a stochastic selection of the training data at each iteration reducing the training cost. Training cost for GP is O(N^3) with N as the size of training data.

Design

ADAM optimization option will be added to the GaussianProcessHandler class in STM.

Impact

No anticipated impacts to existing objects.