Sockeye

Added
- Writing of the convergence/divergence status to the metrics file and guarding against numpy.histogram's errors for NaNs during divergent behaviour.

Changed
- Dynamic batch sizes: `Translator.translate()` will adjust batch size in beam search to the actual number of inputs without using padding.

Added
- `sockeye.score` now loads data on demand and doesn't skip any input lines

Changed
- Do not compare scores from translation and scoring in integration tests.

Added
- Adding the option via the flag `--stop-training-on-decoder-failure` to stop training in case the checkpoint decoder dies (e.g. because there is not enough memory).
In case this is turned on a checkpoint decoder is launched right when training starts in order to fail as early as possible.

Changed
- Do not create dropout layers for inference models for performance reasons.

Changed
- Revert change in 1.18.72 as no memory saving could be observed.