Start setting up TensorFlow for image analysis

GitHub issue(s):
#240

This PR is mainly to have a discussion of how to proceed with add neural network inferencing for images.
The code sets up a couple functions in Python to use tensorflow to do this following the example from: https://docs.databricks.com/_static/notebooks/tensorflow-and-spark.html
Ideal usage eventually should be images.flatMap(apply_inference_on_batch) where images is the result of ExtractImageDetails
Couple issues I ran into:

• The smaller issue: get_tensor_by_name('softmax:0') results in an error KeyError: "The name 'softmax:0' refers to a Tensor which does not exist. The operation, 'softmax', does not exist in the graph." I'm not yet familiar enough with TensorFlow. When I look this up online, a lot of people have they've encountered the same issue and said they have fixed it but did not post the solution.
• The bigger issue: How to connect the scala and python parts together? The best would be if these python functions can be called from the scala. Otherwise re-implementing ExtractImageDetails in pyspark would be an option too.

Please suggest any ideas how to proceed.

Interested parties

@lintool @ruebot @ianmilligan1

Thanks in advance for your help with the Archives Unleashed Toolkit!