Support OpenAI-compatible backends for generation, embeddings, and reranking

[loopyd]

Problem

QMD is currently centered around local llama.cpp model access.

That makes it harder to use QMD with a local or self-hosted OpenAI-compatible server when a user wants to route:

• query expansion / generation
• embeddings
• reranking

through the same remote service.

There are already requests for remote embeddings, but the broader problem is allowing QMD to work with a compatible server across the whole retrieval flow.

Related issues:

• [feature] Why don't support remote embedding services #403
• Feature request: OpenAI-compatible and Gemini API embedding support #428
• Feature Request: Support external API embedding providers (OpenAI/Ollama compatible) #521

Solution

Add support for an OpenAI-compatible backend that can be configured in QMD and used for:

• generation
• embeddings
• reranking

This should allow users to point QMD at a compatible local or self-hosted server, configure model aliases, and run the normal QMD flow without requiring direct local model access.

Reference

PR #619 implements this feature:
#619