Amendmend to PR #7423

[hotzenklotz]

Proposed changes:
@dakshvar22 To my dismay I discovered that my refactoring reintroduced a bug in my original PR #7423. Unfortunately, it was not captured by any of the unit the tests. I came across it when running a new response selector evaluation today. This PR should fix it.

In short, the default value for a response selector key was wrong. It needs to default to None for all "regular" intent examples or otherwise those won't be filtered out further down the pipeline and crash the sklearn reports.

cc @tmbo

Status (please check what you already did):

• added some tests for the functionality
• updated the documentation
• updated the changelog (please check changelog for instructions)
• reformat files using black (please check Readme for instructions)

[dakshvar22]

Thanks @hotzenklotz. Can you please also post what the output now looks like after running a fresh evaluation on the trained response selector model?

[hotzenklotz]

It looks pretty much the same as before. Only now the response selector labels include their parents intent name as well, e.g. ask_faq/question_1.

[dakshvar22]

Thanks! Can you please also add a unit test which would check if training examples with no response keys are correctly filtered during evaluation?

[hotzenklotz]

Can you please also add a unit test which would check if training examples with no response keys are correctly filtered during evaluation?

There is already a unit test for that: https://github.com/RasaHQ/rasa/blob/1.10.x/tests/nlu/test_evaluation.py#L582

[dakshvar22]

@hotzenklotz I just pushed a changelog to indicate this is a bugfix.