Enable reranking for all vectordb instances in search

[osushinekotan]

1416

Summary

This PR introduces a unified reranker implementation within phi.reranker to support reranking functionality across various vectordb instances, not limited to lancedb. It adds the CohereReranker to allow consistent reranking behavior across different vector databases.

Changes

• Implemented a common reranker under phi.reranker to standardize reranking across vectordb instances.
• Added CohereReranker to phi.reranker.cohere.py, enabling the integration of Cohere’s reranker with any vectordb instance.
• Updated search logic to conditionally apply the reranker, allowing flexible and consistent reranking during searches.
• Added reranking_score field to Document to store scores used in reranking.
• Added 06_agentic_rag_with_reranking.py to cookbook.rag as an example demonstrating RAG (Retrieval-Augmented Generation) with reranking integration.

Verified functionality with CohereReranker specifically on lancedb to ensure consistent behavior.

[ashpreetbedi]

this is an exceptional pr @osushinekotan

[ashpreetbedi]

Remarkable work! I'm going to merge, test and release tomorrow