test(metrics): add overlapping-chunk regression fixtures for ContextualPrecisionMetric (rebased on #2743)

[Ruthwik-Data]

Summary

Adds tests/test_metrics/test_contextual_precision_overlapping_chunks.py — a focused regression test suite for ContextualPrecisionMetric behaviour under overlapping-chunk retrieval scenarios.

This supersedes the test file from PR #2692 and is rebased on the merged changes from PR #2743 (RetrievedContextData._group_retrieval_contexts + WCP formula fix). The fixtures now use the new RetrievedContextData with a source field, which is the correct API post-merge.

Closes #2594 (test coverage aspect).

Motivation

PR #2743 fixed two root causes:

1. Source grouping — _group_retrieval_contexts() merges multiple RetrievedContextData chunks sharing the same source string before LLM scoring, preventing redundant-chunk penalisation.
2. WCP formula — _calculate_score() now correctly implements weighted cumulative precision (binary verdicts, rank-discounted).

Without regression tests, these fixes can silently regress. This PR adds four targeted tests that each guard one failure mode from issue #2594.

What this PR adds

Test fixtures

Fixture	Scenario
overlapping_narrative_chunks	Two same-source 20%-overlap 10-K narrative chunks
non_overlapping_mixed_chunks	Three chunks from distinct sources (one relevant)
table_cell_overlap_chunks	Adjacent table-row chunks with shared header (same source/page)

Test cases

Test	Guards against
test_same_source_overlap_does_not_penalise	Redundancy treated as irrelevance (regression for #2594)
test_cross_source_chunks_scored_independently	Source grouping incorrectly merging different documents
test_table_boundary_overlap_not_penalised	Table-header overlap lowering score for financial doc RAG
test_wcp_formula_monotone_with_relevant_rank	WCP score not decreasing as relevant chunk rank drops

Design notes

• Tests use RetrievedContextData(content=..., source=...) — the post-feat(contextual-precision): add RetrievedContextData source grouping and fix weighted precision score #2743 API.
• threshold=0.0 in the WCP monotonicity test isolates formula logic from pass/fail cutoffs.
• No new dependencies; runs with any LLM key configured in conftest.py.
• Supersedes PR test(metrics): add overlapping-chunk test fixtures for ContextualPrecisionMetric #2692 (same scope, updated for merged API).

Type of change

• Test (adding missing tests or correcting existing tests)

[vercel]

@Ruthwik-Data is attempting to deploy a commit to the Confident AI Team on Vercel.

A member of the Team first needs to authorize it.