perf(postgres): optimize KV storage upsert using executemany

[wkpark]

Description

This PR optimizes the document ingestion performance for Postgres-based storage by introducing batch processing
for Key-Value (KV) storage operations. By replacing individual INSERT calls in a loop with executemany and implementing a robust sub-batching mechanism.

it addresses performance bottlenecks where small documents (e.g., <300kB) took several minutes to process.

Changes

• PostgreSQLDB.executemany Implementation: Added a dedicated method to the PostgreSQLDB class to support asyncpg's batch execution, ensuring efficient data transmission to Postgres.
• Refactored PGKVStorage.upsert: Updated the upsert logic for all namespaces (TEXT_CHUNKS, FULL_DOCS, LLM_CACHE, etc.) to collect data into tuples and execute them in a single batch.
• Sub-batching (Chunking) Logic: Introduced a mechanism to split large data sets into smaller sub-batches (defined by _max_batch_size). This prevents memory spikes and database transaction timeouts when processing massive documents.

Checklist

• Changes tested locally
• Code reviewed
• Documentation updated (if necessary)
• Unit tests added (if applicable)

Additional Notes

The sub-batching unit is currently synchronized with the embedding_batch_num (default 10-50), which provides a balanced load for both the LLM embedding phase and the subsequent database write phase.

This PR and PR summary were created with the help of Gemini, with some additional modifications.

[danielaskdd]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4d3ae4af13

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Stream sub-batches instead of buffering full KV payload

PGKVStorage.upsert now accumulates every row into batch_values and only starts writing inside the sub-batch loop, which means large ingestions hold a second full copy of the payload in memory before the first DB write. In the document path (text_chunks.upsert(inserting_chunks)), this can significantly increase peak RSS and cause OOMs on large inputs, whereas the previous per-row implementation did not buffer all rows at once; the sub-batching logic should flush incrementally as each chunk fills rather than after full accumulation.

Useful? React with 👍 / 👎.