Story 4.8: Golden/Master Testing for RAG Pipeline Outputs

As a developer, I want version-controlled golden files for RAG pipeline stage outputs, So that CI detects unintended changes to embedding vectors, retrieval results, reranker scores, and LLM responses.

FRs: FR59 | Sprint: Growth

Acceptance Criteria:

• Golden files for each RAG pipeline stage (embedding vectors, retrieval results, reranker scores, LLM responses) stored in tests/golden/
• pytest test suite compares current pipeline output against golden files
• Diff output shows exact changes when golden mismatch detected
• Golden files updated via explicit flag (not auto-updated)
• Tests run only when AI layer code changes (path-based CI trigger)
• JUnit XML reports collected as artifacts

NFRs: NFR6, NFR8, NFR23, NFR24