Javacpp+dataframes

[robertnishihara]

This is a big PR, and it changes the API. Primarily:

• We're switching from reading from RDDs to DataFrames (Switch from RDDs to DataFrames #61).
• We're switching from wrapping Caffe ourselves to using JavaCPP to wrap Caffe (Switch to using javacpp-presets #60).
• It also includes a number of cleanups and simplifications to the code.

Some things are still incomplete:

• Building on operating systems other than Linux may not work yet (see More of a question than an issue.. what IDE/env is used by committers to do SparkNet development? #64).
• GPUs are disabled (due to Getting Caffe to use the GPU bytedeco/javacpp-presets#147). Because of this, training will be extremely slow. Debugging this is a priority right now. We felt that it didn't make sense to wait for that bug fix to push this change because we are changing the API.