Description

This PR wires the existing rerank client into hierarchical retrieval so THINKING mode can actually use rerank scores
during both starting-point selection and recursive child retrieval.

It also makes rerank failures non-fatal by falling back to vector scores, and fixes the rerank client request path/
response handling to make the integration usable in practice.

Type of Change

• Bug fix (non-breaking change that fixes an issue)
• New feature (non-breaking change that adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Documentation update
• Refactoring (no functional changes)
• Performance improvement
• Test update

Changes Made

• Initialize RerankClient in HierarchicalRetriever when rerank config is available.
• Apply rerank scores in RetrieverMode.THINKING for both global starting points and recursive child retrieval.
• Add safe fallback logic so invalid rerank responses, request failures, or score length mismatches fall back to
  vector scores.
• Update RerankClient to use https:// and return None on invalid/failed rerank calls.
• Add focused retrieval tests covering rerank initialization, THINKING-mode rerank behavior, fallback behavior, and
  QUICK-mode bypass.
• Update the English retrieval documentation to clarify rerank fallback behavior.

Testing

Validated locally with Python 3.11 on Windows.

• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes
• I have tested this on the following platforms:
  • Linux
  • macOS
  • Windows

Checklist

• My code follows the project's coding style
• I have performed a self-review of my code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• Any dependent changes have been merged and published

Additional Notes

• Rerank is only applied in RetrieverMode.THINKING; QUICK mode remains unchanged.
• If rerank returns invalid data or the API call fails, retrieval falls back to vector scores to preserve retrieval
  availability.