Tensorflowonspark

- add grace period to `TFCluster.shutdown()`
- add wide & deep example (contributed by crafet)
- update mnist/pipeline examples to `tf.data`, add instructions, and misc code cleanup (from yileic)
- parameterize versions in pom.xml and code cleanup (from tmielika)
- update Scala Inferencing pom.xml to latest tensorflow-hadoop artifact (contributed by psuszyns)

- Add keras/estimator example
- Update original keras example to latest ` tf.keras` apis
- Update Scala Inferencing pom.xml to latest TF java version
- Allow PS to use CPU on TF-GPU builds (contributed by dratini6)
- More pep8
- More py2/py3 compat

- support [tf.estimator.train_and_evaluate()](https://www.tensorflow.org/api_docs/python/tf/estimator/train_and_evaluate) API
- use local file instead of ppid to uniquely identify executors
- surface GPU allocation errors more readily
- add sharding, epochs, and shuffling to mnist Dataset example.
- TFoS example for criteo data (contributed by amantrac)
- use `tf.train.MonitoredTrainingSession` in mnist/spark example (contributed by wuyifan18)

- Error handling for TF exceptions in InputMode.SPARK (from eordentlich).
- Add timeout for reservations (from eordentlich).
- Errors will exit Spark job with non-zero exit code.
- Fix regression in pipeline API.
- Model export tool

- Added [Scala Inference API](https://github.com/yahoo/TensorFlowOnSpark/wiki/Scala-Inference-API)
- Support driver-side PS node(s) (contributed by winston-zillow)