feat: add ArcadeDB document store integration

[lvca]

Summary

• Add ArcadeDBDocumentStore implementing the full Haystack DocumentStore protocol (count, filter, write/upsert/skip, delete) with automatic database, schema, and HNSW vector index initialization
• Add ArcadeDBEmbeddingRetriever pipeline component for vector similarity retrieval with FilterPolicy support (REPLACE/MERGE)
• Pure HTTP/JSON API integration via requests — no special database drivers needed

ArcadeDB is an open-source multi-model database (~4k GitHub stars) that combines document storage, HNSW vector search (LSM_VECTOR), full-text search, and graph capabilities in a single engine. For Haystack users, this means one database replaces separate backends for document storage and vector retrieval.

Key design decisions

• HTTP-only: All operations use POST /api/v1/command/{database} with SQL — no Bolt protocol, no neo4j driver, no custom binary protocol
• LSM_VECTOR index: Uses ArcadeDB's ACID-compliant HNSW implementation with configurable dimension and similarity metric (cosine, euclidean, dot product)
• ARRAY_OF_FLOATS property type: Required by ArcadeDB's vector index (not generic LIST)
• Post-filtering for vector search: vectorNeighbors() returns {record, distance} maps — metadata filters are applied as a post-filter step
• Automatic schema setup: Creates database, vertex type, properties, unique index, and vector index on first use — zero manual setup required

How did you test it?

• 14 unit tests for filter conversion (no ArcadeDB required)
• 12 integration tests against a live ArcadeDB Docker container covering: count, write (overwrite/skip/duplicate), delete, filter (equality/comparison/AND), embedding retrieval, and serialization round-trip
• End-to-end pipeline example with ArcadeDBEmbeddingRetriever demonstrating filtered vector search
• All 26 tests pass against arcadedata/arcadedb:latest

Checklist

• I have read the contributors guidelines and the code of conduct
• I added unit tests and updated the docstrings
• I've used one of the conventional commit types for my PR title: feat:
• CI workflow with ArcadeDB Docker service included
• Labeler configuration updated
• README inventory table updated

[CLAassistant]

All committers have signed the CLA.

[julian-risch]

Thank you for opening this pull request @lvca ! Looks quite good to me already but there are also a couple of things we need to change before we can merge it. I'll make most of the changes myself. What you could please do is open a PR in https://github.com/deepset-ai/haystack-integrations with a page to highlight this new integration using this structure. Feel free to reuse what you had in mind for the README.md there.

[julian-risch]

I ran the example code successfully on my local machine and applied suggestions from my code review. Main task remaining is to use mixin DocumentStore tests based on https://github.com/deepset-ai/haystack/blob/main/haystack/testing/document_store.py#L252 I'll do that next.

[julian-risch]

I updated the tests to use DocumentStore mixin tests. Both unit and integration tests passed locally.