Add reranking document post processor

[KoreaNirsa]

Closes #5903

Overview

This PR adds reusable document reranking support to the RAG module.

The new post-processor supports the common RAG pattern of retrieving a broader candidate set from a vector store, reranking those candidates with a more precise relevance strategy, and passing only the best-ranked documents to generation.

The implementation intentionally does not define the ranking strategy itself. Instead, applications provide a DocumentReranker, which can delegate to an external reranking model or service, use custom metadata-based scoring, apply rule-based ranking, or combine multiple relevance signals.

Changes

• Adds DocumentReranker as a functional contract for reranking retrieved documents based on a Query
• Adds RerankingDocumentPostProcessor to apply a DocumentReranker after retrieval
• Adds optional topN support to keep only the highest-ranked reranked documents
• Adds unit tests for reranking, topN truncation, validation, and empty input behavior
• Updates the RAG documentation with an example using RetrievalAugmentationAdvisor

[cylin0201]

Hi @KoreaNirsa, nice work on the reranking post-processor!

Just a heads-up: PR #5887 (linked to issue #5776, filed on Apr 8) covers similar territory with a ScoringModel abstraction that follows Spring AI's Model<TReq, TRes> pattern, including a Jina AI provider implementation and auto-configuration.

Flagging this for the maintainers' awareness so both approaches can be evaluated together. 🙂