New training data format (Maturation)

[degiz]

Umbrella ticket for the maturation of the New Training Data Format.

We can logically split the work into parts:

Feedback on 2.0:

• NLU: intents should start with - as noted here - Updates in NLU 2.0 format - multi line training examples #6243

Interactive learning:

• NLU: needs new YAML writer - Add schema and writer for 2.0 NLU data #6252
• Core: needs new YAML writer - Add schema and writer for 2.0 Core data  #6374

New features:

• Add version field to the YAML files with training data format version - Add version string to the 2.0 training data format #6246
• Converter tool from 1.0 to 2.0 - Create intitial CLI converter for 2.0 training data #6402

[tmbo]

@degiz is there a ticket for the yaml stories writer (Core)? I guess it will probably be only one writer anyways since writing really is just dumping a dict into yaml - so not really a difference between nlu and core (apart from the preprocessing needed for the blocks that need the nlu example syntax with our artificial -)

[degiz]

@tmbo it's a little bit trickier than just dumping the dict, bc for both the NLU and Core we use internal representation of the training data, TrainingData and List[StoryStep] respectively, they do not represent the data "as in the training data files".
All tickets are in the Issue description.