Ivis

Supervised mode added to ivis. Additional features:

* Add ``classification_weight`` parameter to allow users to tune balance between classification vs. triplet loss.
* Add Ivis callbacks module for ivis-specific callbacks such as checkpointing during training. Ivis object code changed to deal with provided callbacks.
* Tensorboard callbacks
* Sparse matrix support in supervised mode

Significant improvement in processing speed for both ``precompute=True`` and ``precompute=False`` option using Keras Sequence generator. Addresses 21 .

Bug fixes
PyPi release

* Improved model serialisation
* Bug fixes
* Tests

* Training can be restarted if interrupted or migrated from another instance
* Model complete model serialisation and persistence
* Commonly used Deep Neural Network architectures can be set using keyword parameters

Support for parallel retrieval of KNNs
Progress indicators
General bug fixes and code cleanup