changes in ragas_metrics module

[MukulLambat]

Description

This PR updates Giskard’s RAG evaluation to be compatible with the async metrics API introduced in ragas>=0.3.x, while fixing the original RagasLLMWrapper instantiation issue.

In ragas 0.3.x, the API introduced two main breaking changes that affected Giskard:

1. BaseRagasLLM added a new abstract method is_finished(), which caused:

   TypeError: Can't instantiate abstract class RagasLLMWrapper with abstract method is_finished
   

2. Metrics such as AnswerRelevancy, Faithfulness etc. moved to an async, sample-based API (SingleTurnSample + single_turn_ascore / multi_turn_ascore) instead of the old synchronous .score(dict) / .single_turn_score(...) API.

---

What this PR does

1. Fix LLM wrapper instantiation

   • Implements is_finished() in RagasLLMWrapper to conform to the new BaseRagasLLM interface.
   • The implementation inspects generation metadata / finish reasons so that RAGAS can correctly decide when a completion is done.

2. Adopt the async metrics API with SingleTurnSample

   In RagasMetric.__call__:

   • We continue to build the existing ragas_sample dict via prepare_ragas_sample(...).

   • We then construct a SingleTurnSample:

     sample = SingleTurnSample(
         user_input=ragas_sample.get("user_input") or ragas_sample.get("question"),
         response=ragas_sample.get("response") or ragas_sample.get("answer"),
         retrieved_contexts=ragas_sample.get("retrieved_contexts") or ragas_sample.get("contexts"),
         reference=ragas_sample.get("reference") or ragas_sample.get("ground_truth"),
     )

   • Metric evaluation is now done via the async RAGAS API, using an internal coroutine:

     async def _compute_score():
         if hasattr(self.metric, "single_turn_ascore"):
             return await self.metric.single_turn_ascore(sample)
         elif hasattr(self.metric, "multi_turn_ascore"):
             return await self.metric.multi_turn_ascore(sample)
         else:
             raise AttributeError(
                 f"{self.metric} has neither single_turn_ascore nor multi_turn_ascore "
                 "— check ragas version or metric type."
             )
     
     loop = asyncio.get_event_loop()
     val = loop.run_until_complete(_compute_score())
     return {self.name: val}

   • This aligns Giskard with the current RAGAS metric interface, where metrics are evaluated asynchronously over SingleTurnSample (or multi-turn equivalents).

3. Update tests to use AsyncMock

   • tests/rag/test_ragas_metrics.py is updated so that metric mocks reflect the new async API:

     • AsyncMock is used to mock single_turn_ascore in:

       • test_ragas_metric_computation_with_context
       • test_ragas_metric_computation

     • Tests still assert the same observable behaviour: RagasMetric.__call__ returns a dict like {"test": 0.5}.

   • Existing behaviour for the “missing context” case is preserved; the corresponding test continues to check that we return 0.0 and log a warning.

---

Behavioural impact

• RAG evaluation now works again with ragas>=0.3.x, using the async metric API and SingleTurnSample.

• The original error:

  Can't instantiate abstract class RagasLLMWrapper with abstract method is_finished
  

  is fixed by implementing is_finished() in the wrapper.

• Metric computation is driven by single_turn_ascore / multi_turn_ascore, matching the current RAGAS documentation for legacy metrics.

---

Related Issue

This PR fixes #2218.

---

Type of Change

• 🔧 Bug fix (non-breaking change which fixes an issue)
• 🥂 Improvement (non-breaking change which improves an existing feature)

---

Checklist

• I’ve read the CODE_OF_CONDUCT.md document.
• I’ve read the CONTRIBUTING.md guide.
• I’ve written tests (or updated existing ones) for the methods and classes I touched.
• I’ve updated docstrings where necessary.
• I’ve updated pdm.lock by running pdm update-lock (only applicable if pyproject.toml was modified).

[SurajBhar]

@kevinmessiaen Please chek the new PR with updated changes! We have reversed the changes made in build-python.yml, the file is in its original state. Please cross check it and let us know your feedback on this PR!