[FEATURE] Support Actual Reranker Models

[versecafe]

Problem Statement

Support for dedicated rerank models that return numerical thresholds such as cohere rerank 3.5 and voyage rerank 2 as they tend to be more reliable compared to using traditional llms and vector search.

Proposed Solution

A wrapper over rerank models given that the vercel AI SDK does not provide one, they share a relatively standard interface so wrapping them isn't very difficult.

Component

RAG

Alternatives Considered

Just creating a custom rerank() for my own projects and matching the return type signature.

Example Use Case

Same usage as the current rerank system just changing the function params of rerank()

Additional Context

No response

Verification

• I have searched the existing issues to make sure this is not a duplicate
• I have provided sufficient context for the team to understand the request