Draft: Merge v1.0 from develop

[kengelbrecht]

This includes a lot of refactorings and cleanup steps of the entire code, leading to some changes in how the components are named and used.

Refactored NLUdatset

• Renamed NLUdataset to NluDatset
• removed out_of_scope argument to constructor; intents list will never include OUT_OF_SCOPE_TOKEN
• removed seed argument to constructor as seed should be controllable with more flexibility
• some previously public properties are now more controlled (private or read-only)
• to_records() is deprecated, use to_json() instead
• refactored method to_json
  • removed argument records (as it was probably for a very specific use case we do not have anymore)
  • returns a list of dicts (like to_records()) instead of a JSON string now
• refactored method dataloaders.utils.from_json to match the logic of to_json(). One feature that is not symmetric is that only utterance, intent, and entities will be imported. This is to avoid having changing arguments depending on what properties a dataset can have. Also, we will probably never use this method to load data with more attributes.
• NluDataset.getitem() returns a dataset instead of a record from _data when key is a numbers.Integral
• NluDataset.sample(): Removed default value for size, default for random_state set to None, argument stratification renamed to stratify for consistency with train_test_split and sklearn
• removed argument shuffle from method subsample_by_intent_frequency. Can use ds.suffle().subsample_by_intent_frequency(). The current implementation never did anything useful with the shuffle argument.
• changed default values of method train_test_split() (now inherited from sklearn as None)

Refactored vendors

• removed scikit-learn methods like fit(), predict(), score()
• removed Sklearn superclasses (thsi feature was never really tested)
• made alias an instance attribute
• made all vendor attributes private (previously attributes were generally public even if meant for internal use)

Renamed classifiers

Renamed some modules

• datasets -> nlu_dataset
• vendors.vendors -> vendors.vendor
• modules with vendor implementations renamed to match class names

Moved dataset loaders to nlubridge.dataloaders subpackage

• names include the vendor now
• can be imported from top level module