[Agentic Search] Add reranking support for agentic search pipelines

[ohltyler]

Description

Agentic search now supports reranking, allowing retrieved results to be reordered using a reranking model or field-based scoring for improved relevance and precision. The frontend should allow users to configure a rerank search response processor in the agentic search pipeline.

Current Behavior

The agentic search UI creates a search pipeline with only the agentic_query_translator request processor. There is no way to add a rerank response processor through the UI.

Expected Behavior

• When configuring an agentic search workflow, users should be able to optionally add a rerank response processor to the search pipeline
• Support at least two reranking modes:
  • by_field: Rerank results based on a numeric field value (user selects the target field)
  • ml_opensearch: Rerank results using a cross-encoder model (user selects a deployed cross-encoder model and document fields)
• The UI should show the reranking configuration as an optional section in the pipeline configuration
• The exported pipeline should include both the agentic_query_translator request processor and the rerank response processor

Backend Reference

The agentic search pipeline supports adding a rerank response processor alongside the agentic_query_translator request processor. Both by_field and ml_opensearch reranking modes are supported.

References

• Reranking agentic search results
• Rerank processor
• Meta issue: [META] Agentic Search Frontend Feature Parity with 3.6 Backend Enhancements #865

[andrross]

Catch All Triage - 1 2