fix(retrievers): apply retriever_weights in reciprocal_rerank_fusion mode

[VANDRANKI]

Summary

Setting retriever_weights on a QueryFusionRetriever with mode="reciprocal_rerank" has no effect — all retrievers are weighted equally. The weights are silently ignored.

Root cause

_reciprocal_rerank_fusion iterates over results.values(), discarding the (query_str, retriever_idx) key. Each node always gets 1.0 / (rank + k) regardless of its retriever's weight:

# Before (broken)
for nodes_with_scores in results.values():  # key discarded
    ...
    fused_scores[hash] += 1.0 / (rank + k)  # no weight applied

_relative_score_fusion already does this correctly with retriever_idx = query_tuple[1] and self._retriever_weights[retriever_idx].

Fix

Switch to results.items(), extract retriever_idx from the key, and scale the RRF contribution by self._retriever_weights[retriever_idx]:

for query_tuple, nodes_with_scores in results.items():
    retriever_idx = query_tuple[1]
    weight = self._retriever_weights[retriever_idx]
    ...
    fused_scores[hash] += weight / (rank + k)

This mirrors the weight application in _relative_score_fusion and was introduced in PR #11667 but not carried through to the RRF path.

Fixes #21444

[gautamvarmadatla]

hi @VANDRANKI there is already a PR i created along with the issue #21445. Please review existing work before mass opening AI PRs :)

Infact, you even commented on my PR saying LGTM lol

[VANDRANKI]

Closes as duplicate of #21445 by @gautamvarmadatla, which was already open before this PR was created. Apologies for the noise - I should have checked for existing open PRs before opening this. To make it worse, I had already reviewed #21445 and left LGTM on it, then opened this competing PR. That was a serious oversight. Please disregard this PR and consider #21445 instead.