* Initial checkin to rebase to Initialziers to pick up changes to ndarry Shape

* Initial Checkin for losses

* Fix reshape in sparseCategoricalCrossentropy()

* Apply various fixes to JavaDoc

* Change Tuple to LossTuple

* Repair JavaDOx

* Fixed AllAxis to hanlde dynamic shape when static shape rank is unknown.

* change method name allAxis to allAxes

* change private method binaryCrossentropy to binaryCrossentropyHelper

* Fixed squeezeOrExpandDimensions to make sure the updated labels, predictions and weights are returned in LossTuple

* Fix JavaDoc,
Add in rangeCheck and valueCheck
Misc fixes based on review

* Fix unused imports and add @SuppressWarnings("unchecked") for casts.

* Add copyright

* Add CastHelper and used that for all casts

* Fix JavaDoc, change snake case to camel case.

* Change class LossesImpl to LossesHelper

* Remove commented out JavaDoc

* Changed method name from smoothLabelsBinaryX to smoothBinaryLabels,
smoothLabelsCatX to smoothCategoricalLabels.

Added clarification oin JavaDoc for cosineSimilarity to describe the difference between the mathematical definition for cosine similarity and the loss definition.

* Fixed JavaDoc for labelSmoothing

* Fixed JavaDoc to change label_smoothing to labelSmoothing.

* Fix formatting

* replace label_smoothing with labelSmoothing.
fix typo error in JavaDoc comment

* Add copyright to test cases

* Fix copyright to attribute TensorFlow Authors.

* Fix typo on broadcast in JavaDoc

* Fix typo on broadcast in JavaDoc