[Test]: cover RAG preparation and search helpers

[viiccwen]

Summary

Add deterministic unit coverage for RAG data loading, document preparation, BM25 search, hybrid ranking, and source hashing. Tests isolate module state and avoid embedding or FAISS initialization.

Related Issue

Fixes #86

Type of Change

• Bug fix
• New feature
• Refactor
• Documentation
• Configuration or deployment
• Tests

Changes

• Cover valid, missing, and malformed JSONL inputs plus valid and missing Markdown files using tmp_path.
• Cover FAQ, paragraph, entity, and Markdown document preparation with deterministic keyword extraction.
• Cover BM25 index construction and ranking plus hybrid ranking with in-memory documents and fake semantic results.
• Cover source hashes for present and missing inputs and isolate loader caches and module-level BM25 state between tests.

Testing

• Local tests or checks pass: make test (182 passed)
• Manual verification completed: targeted RAG tests, make precommit, and git diff --check
• Not tested; reason:

Impact Checklist

• Discord bot behavior checked, if affected
• MCP server behavior checked, if affected
• Webhook behavior checked, if affected
• Docker or compose changes checked, if affected
• Environment variables documented, if changed
• Logs do not expose secrets or sensitive data

Notes for Reviewers

This is a test-only change. It does not call initialize_rag_system(), instantiate embedding models, or build/load FAISS indexes. Existing dependency deprecation warnings are unchanged.