GH-5776: Add Jina AI Scoring (Reranker) Module

[cylin0201]

GH-5776: Add Jina AI Scoring (Reranker) Module

Motivation

Spring AI currently lacks a built-in scoring/reranking abstraction. This PR introduces a ScoringModel abstraction in spring-ai-model and provides the first implementation using the Jina AI Reranker API.

Reranking is a critical component in RAG pipelines — it re-scores retrieved documents against the user query to improve relevance ordering before passing them to the LLM.

Changes

1. Core Scoring Abstraction (spring-ai-model)

• ScoringModel, ScoringOptions, ScoringRequest, ScoringResponse, ScoringResult
• ScoringOptions.getModel() added as a default method for backward compatibility
• Registered SCORING in SpringAIModels and SpringAIModelProperties

2. Jina AI Implementation (models/spring-ai-jina)

• JinaScoringApi — REST client with Builder pattern for Jina Reranker API (POST /v1/rerank)
• JinaScoringModel — ScoringModel implementation with retry support, options merging, input validation
• JinaScoringOptions — model, topK, returnDocuments, truncation configuration
• Supported models: jina-reranker-v2-base-multilingual, jina-reranker-v1-*, jina-colbert-v1-en, jina-colbert-v2, jina-reranker-m0, jina-reranker-v3

3. RAG Integration (spring-ai-rag)

• ScoringDocumentPostProcessor — post-retrieval reranking pipeline for RetrievalAugmentationAdvisor

4. Auto-Configuration (auto-configurations/models/spring-ai-autoconfigure-model-jina)

• JinaScoringAutoConfiguration — conditional bean registration
• JinaScoringProperties — spring.ai.jina.scoring.* configuration properties
• Spring Boot Starter (spring-ai-starter-model-jina)

5. BOM Registration

• spring-ai-jina, spring-ai-autoconfigure-model-jina, spring-ai-starter-model-jina added to spring-ai-bom

Test Coverage

Test Class	Type	Count	Notes
JinaScoringModelTests	Unit (MockWebServer)	5	success, merge options, error, empty docs, retry
JinaScoringApiIT	Integration	3	basic rerank, topN, usage stats
JinaScoringModelIT	Integration	4	full model call, topK override, metadata preservation
ScoringDocumentPostProcessorIT	Integration	1	RAG pipeline end-to-end
JinaScoringAutoConfigurationIT	Integration	1	auto-config bean registration

Integration tests require JINA_API_KEY environment variable and are skipped by default.

Usage Example

// Direct usage
JinaScoringApi api = JinaScoringApi.builder()
    .apiKey("jina_...")
    .build();

JinaScoringModel scoringModel = JinaScoringModel.builder()
    .jinaScoringApi(api)
    .options(JinaScoringOptions.builder()
        .model("jina-reranker-v2-base-multilingual")
        .topK(3)
        .build())
    .build();

ScoringResponse response = scoringModel.call("What is Spring AI?", documents);

// With Spring Boot auto-configuration
// application.yml:
// spring.ai.jina.scoring.api-key: jina_...
// spring.ai.jina.scoring.options.model: jina-reranker-v2-base-multilingual

Checklist

• All unit tests pass (mvnw clean verify -pl models/spring-ai-jina)
• Auto-configuration module builds successfully
• Checkstyle / spring-javaformat: 0 violations
• NullAway: all null-safety annotations applied
• DCO Sign-off on all commits
• No unrelated changes included

Closes #5776