debezium/dbz#2040 feat: add Chroma adapter and end-to-end sync pipeli…

[KMohnishM]

Description

Fixes debezium/dbz#2040.

This pull request completes the Week 3 deliverables by introducing the Chroma Vector Store Adapter (ChromaAdapter) and adding the first suite of end-to-end pipeline integration tests to verify real-time event synchronization.

This PR builds directly on top of the SyncManager introduced in PR #7 (gsoc-week-3-sync).

Key Changes

1. Chroma Vector Store Adapter (pydebeziumai/adapters/chroma.py):

   • Implements the ChromaAdapter which inherits from our base VectorStoreAdapter interface.
   • Wraps langchain_chroma.Chroma to enable vector operations inside the SyncManager.
   • Implements upsert and delete operations for document lifecycle sync.
   • Exposes as_retriever() which hooks directly into LangChain's retrieval chains.

2. E2E Sync Integration Tests (tests/integration/test_chroma_pipeline.py):

   • Validates the complete CDC-to-Vector-Store pipeline end-to-end using an ephemeral, in-memory Chroma instance and FakeEmbeddings.
   • Covers the following scenarios through the entire flow (Event Ingestion ➔ Document Builder ➔ SyncManager ➔ Chroma Vector Store):
     • Insert/Snapshot (c/r): Verifies documents are embedded and upserted.
     • Update (u): Verifies documents are updated (deleted and re-upserted to prevent duplicate keys).
     • Hard Delete (d with soft_delete=False): Verifies documents are fully deleted from the database.
     • Soft Delete (d with soft_delete=True): Verifies document is retained in the vector store but its metadata is updated with _is_deleted=True and its original state is preserved.

Verification Results

All checks passed locally in the WSL testing environment:

• Ruff formatting & linting: Checked and passed (All checks passed!).
• MyPy strict type-checking: Passed (Success: no issues found in 28 source files).
• PyTest test suite: All 56 unit and integration tests passed successfully, including the 4 new integration tests for the Chroma adapter pipeline.

[vjuranek]

LGTM

[KMohnishM]

LGTM

@vjuranek I will update the other 2 PRs and rebase this. Thank you for the review !

[vjuranek]

Merge, thanks @KMohnishM !