Return response-level scores for CounterfactualGenerator and AutoEval

[dylanbouchard]

Is your feature request related to a problem? Please describe.
Currently, CounterfactualGenerator and AutoEval return a dictionary containing only metric values. In contrast, ToxicityMetrics and StereotypeMetrics give users the option to return response-level scores for a more detailed investigation into the assessment results. The request here is to give users the option to return analogous, response-level scores for CounterfactualGenerator, and return response-level toxicity, stereotype, and counterfactual scores with AutoEval.

Describe the solution you'd like
Give users option to return response-level scores with CounterfactualGenerator and AutoEval in returned dictionary. For consistency with ToxicityMetrics and StereotypeMetrics, this should be achieved by adding a new boolean argument called return_data to CounterfactualGenerator.evaluate and AutoEval.evaluate. The dictionary format returned should be consistent across these classes: a "metrics" key with metric values, and a "data" key with responses and response-level scores.

Additional context
Along with counterfactual.py, auto.py, demo notebooks, unit tests, and readme should also be updated.

[dylanbouchard]

@mohitcek is assigned to this issue. The requested counterfactual updates are addressed in PR #66

[dylanbouchard]

@mohitcek resolved this issue. Changes are reflected in v0.3.0