Fix OpenSearch index initialization and align knn parameters

[CodeVishal-17]

Summary

Fixes an issue where OpenSearch indices (e.g. documents) were not created
deterministically, causing search or ingestion to fail with
index_not_found_exception when called before onboarding completed.

Changes

• Ensure required OpenSearch indices exist before executing search queries
  by calling _ensure_opensearch_index() in search_service
• Align ef_construction kNN parameter between backend and Langflow
  OpenSearch component to prevent mapping conflicts

Fixes

• [Bug]: Data ingestion fails due to OpenSearch "documents" index not found exception #845

Testing

• Verified that search now triggers index creation on a fresh startup
  without completing onboarding
• Note: Full backend Docker build fails on Windows due to CUDA/PyTorch
  dependency download timeouts; this is unrelated to this change

[CodeVishal-17]

CI failure appears to be due to a network error while downloading
Docling/EasyOCR models during the backend Docker build
(uv run docling-tools models download), which is unrelated to this change.
No application code paths modified by this PR affect that step.