Implement the equivalent of Ollama CLI's --hidethinking flag specifically for the "Replace current selection" action. This would:
- Let the model think internally (better quality responses)
- Only insert the final answer without <think></think> tags
- Keep the replacement clean and ready to use
