Add Amazon Bedrock Knowledge Base VectorStore implementation

Expected Behavior

Add a VectorStore implementation for Amazon Bedrock Knowledge Bases that allows similarity search against pre-configured Knowledge Bases using the Bedrock Agent Runtime Retrieve API.

BedrockKnowledgeBaseVectorStore vectorStore = BedrockKnowledgeBaseVectorStore
    .builder(client, "knowledge-base-id")
    .topK(5)
    .similarityThreshold(0.5)
    .build();

List<Document> results = vectorStore.similaritySearch(
    SearchRequest.builder().query("What is the return policy?").build());

Current Behavior

Spring AI supports Bedrock for chat and embedding models, but there's no integration with Bedrock Knowledge Bases. Users must implement their own integration or use a separate vector store even when their documents are already in a Bedrock Knowledge Base.

Context

Bedrock Knowledge Bases provide fully managed RAG with automatic document ingestion, chunking, and embedding. Unlike other vector stores, it doesn't require an EmbeddingModel - the Knowledge Base handles embeddings internally.

Key features needed:

• Similarity search with topK and threshold
• Metadata filtering (Spring AI filter expressions → Bedrock RetrievalFilter)
• SEMANTIC and HYBRID search types
• Optional reranking model support
• Support for multiple data sources (S3, Confluence, SharePoint, etc.)

Note: This is read-only - add() and delete() throw UnsupportedOperationException since documents are managed via KB data source sync.

Start by reviewing Spring AI's existing Bedrock chat and embedding integrations, then trace the VectorStore and SearchRequest contracts alongside the Bedrock Agent Runtime Retrieve API. The implementation is done when read-only similarity search supports topK, thresholds, metadata filters, SEMANTIC or HYBRID search, optional reranking, and rejects add() and delete() as specified.