Fixes and adjustment in rag metrics and related inference engines

[lilacheden]

1. add new classification engines and remove obsolete models
2. propagate score_prefix from metricPipeline to its metric
3. Adjust autorag metrics to unitxt flow with rag.response_generation task
4. Add AzureOpenAIInferenceEngine and make rag llmaj use it

[yoavkatz]

Lilach. Thank you for this work. To complete the work, I think we should have metric links and metrics lists (let's think of the names, I'm not sure about them).

metrics.rag.end_to_end.recommended_cpu_only:
metrics.rag.end_to_end.recommended_cpu_only.faithfulness -> link to recommended metric
metrics.rag.end_to_end.recommended_cpu_only.context_relevance -> link to recommended metric
metrics.rag.end_to_end.recommended_cpu_only.answer_correctness -> link to recommended metric
metrics.rag.end_to_end.recommended_cpu_only.context_correctness -> link to recommended metric
metrics.rag.end_to_end.recommended_cpu_only.answer_relevance -> link to recommended metric

metrics.rag.end_to_end.recommended_local_llm (can we think of a better name?)
...
metrics.rag.end_to_end.recommended_llm_as_judge
....

And then update the example to use the new metrics.

unitxt/examples/evaluate_rag_end_to_end_dataset_with_given_predictions.py

[coveralls]

coverage: 80.088% (+0.03%) from 80.057%
when pulling 056a313 on metrics_fix
into 21482b9 on main.