feat: Hybrid Search with BM25 + BGE-M3 Semantic Retrieval, RRF Fusion and BGE Reranker

[Parth-2701]

Overview

This PR implements a hybrid retrieval pipeline for the DevRev Search benchmark
that combines lexical and semantic search, fuses results using Reciprocal Rank
Fusion (RRF), and reranks final candidates using a cross-encoder.

What Changed

• Replaced single-stage dense retrieval with a dual retrieval system
  combining lexical and semantic search
• Introduced a rank fusion strategy to intelligently merge and deduplicate
  results from both retrievers
• Added a neural cross-encoder reranking stage as the final step for
  precise relevance estimation
• Refactored the entire search module into clean, modular, independently
  testable components

Models Used

• Embedding: BAAI/bge-m3 (1024-dim, multilingual)
• Reranker: BAAI/bge-reranker-v2-m3

[Parth-2701]

@nimit2801 and @prakhar7651 Please have a look at the PR description validation.

[prakhar7651]

Hey!
These are your scores.
Recall@10: 0.3158
Precision@10: 0.2973