Describe the bug

OpenSearchDocStatusStorage.upsert() calls _run_chunked_async_bulk(..., raise_on_error=False) but previously discarded the returned failed collection. OpenSearch bulk requests can complete at the request level while individual items fail, so a DocStatus write can be treated as successful even when a single item reports a mapping error, retryable status, or other per-item failure.

DocStatus records are LightRAG pipeline bookkeeping for document processing states. If a DocStatus row is not actually written but the pipeline continues as if it was, a document can lose the reliable record of whether it is pending, failed, or processed.

Steps to reproduce

1. Use the OpenSearch DocStatus backend.
2. Make the OpenSearch bulk helper return a response-level success with item-level failures, for example:

d1 -> status=400 mapper_parsing_exception
d2 -> status=503 transient failure

3. Call OpenSearchDocStatusStorage.upsert() for those document ids.

Before PR #3354, upsert() ignored the returned failed list and completed without surfacing either item failure.

Expected Behavior

OpenSearchDocStatusStorage.upsert() should inspect per-item bulk failures and fail fast when any DocStatus write was not accepted by OpenSearch.

LightRAG Config Used

OpenSearch DocStatus storage backend. The failure mode is independent of a specific deployment config because it follows from OpenSearch bulk response semantics.

Logs and screenshots

No live OpenSearch service is required to demonstrate the bug. PR #3354 adds mocked regression coverage for the bulk helper returning item-level failures.

Additional Information

• LightRAG Version: main before PR fix(opensearch): handle DocStatus bulk item failures #3354
• Operating System: not OS-specific
• Python Version: not Python-version-specific
• Related Issues: PR fix(opensearch): handle DocStatus bulk item failures #3354