fix: apply retriever weights in reciprocal rank fusion

[fengjikui]

Summary

• Fixes [Bug]: QueryFusionRetriever ignores retriever_weights when mode="reciprocal_rerank" #21444.
• Applies retriever_weights when QueryFusionRetriever uses reciprocal rank fusion.
• Adds a regression test covering retriever_weights=[1.0, 0.0] with mode=FUSION_MODES.RECIPROCAL_RANK.

Root cause

_reciprocal_rerank_fusion() iterated over results.values(), which discarded the (query, retriever_idx) key. Without retriever_idx, the reciprocal rank score always used an implicit equal weight.

Fix

The method now iterates over results.items(), reads the retriever index from the key, and scales each reciprocal-rank contribution by self._retriever_weights[retriever_idx].

Validation

• uv run --frozen --project llama-index-core pytest llama-index-core/tests/retrievers/test_fusion_retriever.py -q
• uv run --frozen --project llama-index-core ruff format --check llama-index-core/llama_index/core/retrievers/fusion_retriever.py llama-index-core/tests/retrievers/test_fusion_retriever.py
• uv run --frozen --project llama-index-core ruff check llama-index-core/llama_index/core/retrievers/fusion_retriever.py llama-index-core/tests/retrievers/test_fusion_retriever.py

AI assistance

I used AI assistance to help trace the narrow code path and draft this small patch, then manually reviewed the diff and validated it with the commands above.

[gautamvarmadatla]

Hi @fengjikui, please check existing open PRs before mass opening new ones. Most of your PRs are duplicating existing work, and I’ve already left comments on multiple other PRs of yours. Even for this PR there is #21445 :)

[fengjikui]

Thanks for pointing this out. I re-checked #21445 and confirmed it already covers the same #21444 issue, the same QueryFusionRetriever reciprocal-rank weighting path, and the same test area. Closing this PR as duplicate so review can stay focused on the earlier PR.