Support local Qwen VLM reranking for search results

[Binary67]

Summary

• Add a local Qwen3-VL Instruct reranker for sentrysearch search --rerank when searching local Qwen indexes.
• Share rerank prompt construction, response schema, JSON parsing, and fenced JSON handling across Gemini and local VLM rerankers.
• Route reranking by backend: local searches use the Qwen reranker, while Gemini and qwen-cloud searches continue using Gemini Flash.
• Preserve embedding-ranked results when reranking cannot load or score candidates.
• Update CLI behavior, README docs, and unit tests for local reranking.

Testing

• uv run --group test pytest tests/test_reranker.py tests/test_qwen_reranker.py tests/test_gemini_reranker.py tests/test_cli.py -q
• git show --check --no-color HEAD

Related Issue

Related to #71