Koog storage backend (kdrant-koog)

[tonytonycoder11]

A Kdrant-backed Koog storage, so a Koog RAG agent can use Qdrant.

The original blocker has cleared: ai.koog:rag-vector-jvm:1.1.1-beta and its dependencies
rag-base-jvm:1.1.1 and embeddings-base-jvm:1.1.1 all resolve on Maven Central now.

What it also turned up is that this issue named the wrong interface. VectorStorageBackend<Document>
has no search method: Koog's EmbeddingStorage ranks by streaming allDocumentsWithPayload() from the
backend and scoring in memory. A Kdrant implementation of it would make every query pull the whole
collection over HTTP and rank it client-side, which is worse than not using a vector database at all,
since it pays for the index and then ignores it.

The right target is SearchStorage<Document, SimilaritySearchRequest>, where search is the
implementation's own and Qdrant does the work. SimilaritySearchRequest carries queryText, limit,
offset, minScore and a filterExpression, which maps onto Kdrant's search DSL the way the Spring AI
and LangChain4j adapters already do.

Two things to settle before building it:

• SearchStorage lives in rag-base, which is stable at 1.1.1, while rag-vector is published
  only as 1.1.1-beta. Targeting the search interface may avoid depending on a beta at all, which
  matters for a module Kdrant would publish under its own stability policy.
• The request carries a text query rather than a vector, so the module needs an embedder, as the other
  framework adapters do.