Can Reasoning Models Reason about Hardware? An Agentic HLS Perspective