fix: use async gptel-request with dedicated Ollama backend

[barbed-chili-glue]

Summary

• Replace synchronous gptel-request calls with async callback pattern
• Capture Ollama backend from existing gptel config into gptel-ollama-backend variable
• Pass :backend explicitly to gptel-request to avoid routing to Gemini/Together AI
• Remove duplicate Ollama backend definition

Problem

The proofreading functions used gptel-request synchronously (treating the return value as a string) and did not specify :backend. This caused two errors:

1. The request went to whatever gptel-backend was currently active (Gemini), not Ollama
2. gptel-request is asynchronous — the return value is an FSM state object, not a string

Fix

• gptel-ollama-backend is now set in the gptel :config block by capturing the return value of gptel-make-ollama
• Both proofreading functions use :backend gptel-ollama-backend and :callback for async response handling