Fix errors during training in ResponseSelector

[tabergma]

Proposed changes:

• Fix errors during training in ResponseSelector

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)

[tabergma]

It seems like that there is more to it. We got the error

Number of examples differs for keys 'dict_keys(['text_features', 'intent_features', 'intent_ids', 'text_mask', 'intent_mask'])'. Number of examples should be the same for all data in session data.

for the following pipeline:

pipeline:
- name: WhitespaceTokenizer
  intent_split_symbol: _
- name: RegexFeaturizer
- name: CRFEntityExtractor
- name: EntitySynonymMapper
- name: CountVectorsFeaturizer
  analyzer: word
  min_ngram: 1
  max_ngram: 1
- name: CountVectorsFeaturizer
  analyzer: char_wb
  min_ngram: 1
  max_ngram: 4
- name: ConveRTFeaturizer
- name: EmbeddingIntentClassifier
  epochs: 151
  embed_dim: 20
- name: ResponseSelector
  epochs: 100

Training of ResponseSelector fails.

Still debugging

[tabergma]

@dakshvar22 Thanks for your help! Ready for another review.

[dakshvar22]

Looks good. Tested it on carbon bot data with cross fold validation.
Added a test for capturing any errors like this in the future.