Sockeye

Changed
- Changed to a custom speedometer that tracks samples/sec AND words/sec. The original MXNet speedometer did not take
variable batch sizes due to word-based batching into account.

Fixed
- Fixed entry points in `setup.py`.

Changed
- Update to [MXNet 1.0.0](https://github.com/apache/incubator-mxnet/tree/1.0.0) which adds more advanced indexing
features, benefitting the beam search implementation.
- `--kvstore` now accepts 'nccl' value. Only works if MXNet was compiled with `USE_NCCL=1`.

Added
- `--gradient-compression-type` and `--gradient-compression-threshold` flags to use gradient compression.
See [MXNet FAQ on Gradient Compression](https://mxnet.incubator.apache.org/versions/master/faq/gradient_compression.html).

Fixed
- Taking the BOS and EOS tag into account when calculating the maximum input length at inference.

Fixed
- fixed a problem with `--num-samples-per-shard` flag not being parsed as int.

Added
- New CLI `sockeye.prepare_data` for preprocessing the training data only once before training,
potentially splitting large datasets into shards. At training time only one shard is loaded into memory at a time,
limiting the maximum memory usage.

Changed
- Instead of using the --source and --target arguments sockeye.train now accepts a
--prepared-data argument pointing to the folder containing the preprocessed and sharded data. Using the raw
training data is still possible and now consumes less memory.