New Features - Added support for Model Quantization with Calibration

• Implemented model quantization by adopting the TensorFlow approach with calibration by borrowing the idea from Nvidia's TensorRT. 
• The focus of this work is on keeping quantized models (ConvNets for now) inference accuracy loss under control when compared to their corresponding FP32 models. 
• Please see the example on how to quantize a FP32 model with or without calibration. (#9552)

New Features - Added Exception Handling Support for Operators

• Implemented Exception Handling Support for Operators in MXNet. 
• Transports backend C++ exceptions to the different language front-ends and prevent crashes when exception is thrown during operator execution. (#9681)

New Features -

Enhanced FP16 support

• Adds support for distributed mixed precision training with FP16. It supports storing of master copy of weights in float32 with the multi_precision mode of optimizers.
• Improved speed of float16 operations on x86 CPU by 8 times through F16C instruction set.
• Added support for more operators to work with FP16 inputs.
• Added a tutorial on using mixed precision with FP16. 

New Features - Added Profiling enhancements

• Built-in profiler enhanced to support native Intel:registered: VTune:tm: Amplifier objects such as Task, Frame, Event, Counter and Marker from both C++ and Python -- which is also visible in the Chrome tracing view. 

• Runtime tracking of symbolic and imperative operators as well as memory and API calls. 

• Tracking and dumping of aggregate profiling data. 

• Profiler also no longer affects runtime performance when not in use.