v0.0.3

• Model prediction API.
• Model checkpoint loading and saving.
• IJulia Notebook example of using pre-trained imagenet model as classifier.
• Symbol saving and loading.
• NDArray saving and loading.
• Optimizer gradient clipping.
• Model training callback APIs, default checkpoint and speedometer callbacks.
• Julia Array / NDArray data iterator.
• Sphinx documentation system and documents for dynamically imported libmxnet APIs.