debezium/dbz#2229 feat: implement batching interfaces for vector stor…

[KMohnishM]

…e adapters and add benchmark script

Description

This PR implements the first part of the GSoC Week 8 milestone for the PyDebeziumAI project. It introduces list-based batch processing interfaces to the VectorStoreAdapter contract and implements them across all three backends (Chroma, PGVector, and Milvus). It also adds a performance profiling utility to establish a baseline and measure batching speedup.

Proposed Changes

• pydebeziumai.adapters.base:
  • Added abstract upsert_batch(self, documents: list[Document]) and delete_batch(self, doc_ids: list[str]) methods to the VectorStoreAdapter interface.
• pydebeziumai.adapters.chroma / pgvector / milvus:
  • Implemented the batching methods using the underlying vector stores' native bulk client operations (e.g., add_documents and delete using array parameters) to reduce network round-trips.
• tools/benchmark.py:
  • Added a benchmark script that runs sequential vs. batched synchronization cycles against an in-memory Chroma instance using synthetic document streams, measuring end-to-end latency and throughput.

Verification Results

Running the benchmark suite (python3 tools/benchmark.py --num-records 100 --batch-size 20) demonstrates a 5.23x throughput speedup:

• Sequential Sync: ~50.88 records/sec (Average latency: 19.65 ms per document)
• Batched Sync: ~266.26 records/sec (Average latency: 75.11 ms per batch of 20)

Unit tests have been updated to ensure all mock adapters implemented in test suites comply with the updated interface.