Revscoring

Added
- Added Icelandic language support.
- Multilabel random forest.
- Added Catalan language assets.
- Fast scoring for fast cross validation.

Changed
- Python 3 is now a hard requirement.
- Row formatting for long label names.
- Support for specifying label weights in multilabel classification.
- Use nltk stopwords library instead of homemade list wherever possible.

Fixed
- Selectors: Count documents, labels per document instance, not per token.
- Fixed label type issue in label-config.

Changed
- Use pytest-cov to collect coverage.
- Ignore test-like functions using tox.ini

Changed
- Migrate from nosetests to pytests.
- Updated nltk download instructions to include additional corpora.
- Handle ModelInfo lookup error.

Fixed
- Key pattern in `format_str` for ModelInfo.

Added
- Additional tests to filters and mappers.

Changed
- Limit memory usage of threshold statistics.
- Use `__slots__` in ScaledClassificationMatrix to reduce memory usage.
- Added `threshold_ndigits` to classification params.

Added
- Added Serbian language assets.

Changed
- Natively handle bz2 compressed model files.

Added
- Bosnian language assets.

Changed
- Changes uk dict recommendation to aspell (from myspell).

Fixed
- Fixed sorting in tune utility.