Sockeye

Fixed
- Transformer models do not ignore `--num-embed` anymore as they did silently before.
As a result there is an error thrown if `--num-embed` != `--transformer-model-size`.
- Fixed the attention in upper layers (`--rnn-attention-in-upper-layers`), which was previously not passed correctly
to the decoder.
Removed
- Removed RNN parameter (un-)packing and support for FusedRNNCells (removed `--use-fused-rnns` flag).
These were not used, not correctly initialized, and performed worse than regular RNN cells. Moreover,
they made the code much more complex. RNN models trained with previous versions are no longer compatible.
- Removed the lexical biasing functionality (Arthur ETAL'16) (removed arguments `--lexical-bias`
and `--learn-lexical-bias`).

Changed
- Updated to [MXNet 0.12.1](https://github.com/apache/incubator-mxnet/releases/tag/0.12.1), which includes an important
bug fix for CPU decoding.

Changed
- Removed dependency on sacrebleu pip package. Now imports directly from `contrib/`.

Changed
- Transformers now always use the linear output transformation after combining attention heads, even if input & output
depth do not differ.

Fixed
- Fixed a bug where vocabulary slice padding was defaulting to CPU context. This was affecting decoding on GPUs with
very small vocabularies.

Fixed
- Fixed an issue with the use of `ignore` in `CrossEntropyMetric::cross_entropy_smoothed`. This was affecting
runs with Eve optimizer and label smoothing. Thanks kobenaxie for reporting.