Closes #19

Summary

This PR adds a small, opt-in retrieval fallback for attribute-style queries (e.g. where/when/born/located/founded) while keeping LinearRAG default behavior unchanged.

What changed

• Added new config fields in src/config.py:
  • enable_hybrid_attribute_fallback (default False)
  • attribute_keyword_boost (default 0.25)
  • attribute_query_keywords (lightweight keyword list)
• Updated retrieval path in src/LinearRAG.py:
  • pass question context into graph retrieval scoring
  • when fallback is enabled and query matches attribute-style intent, apply a small keyword-overlap score boost in passage ranking
• Kept baseline behavior intact when fallback is disabled.

Why

Issue #19 reports hub-entity attribute queries retrieving broad/high-centrality passages while missing relation-specific information. This PR adds a graceful, optional mitigation without changing the paper-faithful default pipeline.

Validation

• python3 -m py_compile src/LinearRAG.py src/config.py src/ner.py src/utils.py
• Manual review of fallback gating path to ensure no behavior change when enable_hybrid_attribute_fallback=False.

If maintainers prefer, I can follow up with a tiny synthetic regression harness in a separate PR to quantify ranking changes on attribute-style examples.