[codex] Add source grouping parity for contextual precision

[cat0825]

Summary

• share the source-based retrieval-context grouping helper used by contextual precision metrics
• apply RetrievedContextData.source grouping to TurnContextualPrecisionMetric before LLM verdict generation, matching the single-turn ContextualPrecisionMetric behavior
• add deterministic source-grouping tests and document source granularity/rank semantics

Why

This is a focused follow-up to #2594 and #2669. The main branch already supports source-based grouping for single-turn contextual precision, but turn contextual precision accepted RetrievedContextData without grouping same-source contexts before verdict generation.

This PR keeps the schema/source-first direction discussed by maintainers, avoids adding a new group_by public API, and supersedes the draft direction from #2729 with tests and docs around the merged source behavior.

Notes

• Plain string retrieval contexts remain independent nodes.
• RetrievedContextData entries with the same source are grouped at the first position where that source appears, preserving best/first-seen rank behavior for fused retrieval lists.
• source can be a section/span ID for section-level scoring or a chunk hash for exact duplicate/cross-arm hybrid search deduplication.

Validation

• PYTHONPATH=. poetry run pytest tests/test_metrics/test_contextual_precision_source_grouping.py -q
• PYTHONPATH=. poetry run pytest tests/test_core/test_imports.py tests/test_core/test_datasets/test_dataset.py -q
• OPENAI_API_KEY= PYTHONPATH=. poetry run pytest tests/test_metrics/test_contextual_precision_metric.py tests/test_metrics/test_turn_contextual_precision_metric.py -q
• PYTHONPATH=. poetry run black --check deepeval/metrics/utils.py deepeval/metrics/contextual_precision/contextual_precision.py deepeval/metrics/turn_contextual_precision/turn_contextual_precision.py tests/test_metrics/test_contextual_precision_source_grouping.py
• PYTHONPATH=. poetry run ruff check deepeval/metrics/utils.py deepeval/metrics/contextual_precision/contextual_precision.py deepeval/metrics/turn_contextual_precision/turn_contextual_precision.py tests/test_metrics/test_contextual_precision_source_grouping.py

Refs #2594

[vercel]

@cat0825 is attempting to deploy a commit to the Confident AI Team on Vercel.

A member of the Team first needs to authorize it.

[cat0825]

CI note: the lint (ubuntu-latest) job is failing on the repository-wide Black check because unrelated existing files would be reformatted. The touched Python files in this PR are formatted; local targeted checks passed:

• PYTHONPATH=. poetry run black --check deepeval/metrics/utils.py deepeval/metrics/contextual_precision/contextual_precision.py deepeval/metrics/turn_contextual_precision/turn_contextual_precision.py tests/test_metrics/test_contextual_precision_source_grouping.py
• PYTHONPATH=. poetry run ruff check deepeval/metrics/utils.py deepeval/metrics/contextual_precision/contextual_precision.py deepeval/metrics/turn_contextual_precision/turn_contextual_precision.py tests/test_metrics/test_contextual_precision_source_grouping.py

I am not formatting the unrelated 58 files in this PR to keep the contribution focused on #2594/#2669.