Simplify CNN model code with library functions

[svrama]

There have been some conventional advances since Fathom's release which lessen the need for things like explicit variable declarations in some of the models (e.g., convolutional networks).

[svrama]

The tf.contrib.slim submodule has been integrated in TensorFlow after Fathom's release. The excellent range of example models now posted by the TensorFlow team demonstrates the use of these wrapper functions for vision models like AlexNet.

The following Fathom models can be refactored using the tf.contrib.slim wrapper functions:

• Residual
• VGG
• AlexNet
• DeepQ (CNN only)