examples: add zero-cloud local hybrid RAG with SQLite FTS5 and embeddings

[Cagrik34]

Summary

Adds a zero-cloud, standalone Hybrid Retrieval-Augmented Generation (RAG) reference example combining embedded SQLite FTS5 (BM25 token search) and Ollama dense embeddings with Reciprocal Rank Fusion (RRF, $k=60$) and grounded streaming citation synthesis.

Key Features

• Zero-Cloud Embedded Dual Retrieval: Utilizes native SQLite FTS5 unicode61 token matching alongside ollama.embed dense vectors, eliminating the need for standalone vector database infrastructure.
• Reciprocal Rank Fusion (RRF, k=60): Resolves score calibration mismatches between cosine similarity and BM25 rank, ensuring robust recall of exact alphanumeric values (policy limits, financial budgets, version tags).
• Streaming Grounded Synthesis: Demonstrates real-time token streaming with ollama.chat(stream=True) and strict citation grounding ([1], [2]).
• Resilient Execution: Includes graceful fallback simulation when the local Ollama daemon is offline, guaranteeing zero unhandled runtime exceptions during automated testing.

Files Added

• examples/hybrid-rag.py
• examples/README.md (Index entry under ### Hybrid RAG)

Verification

Tested and verified with Python 3.10+ and the Ollama Python SDK.