Captures screen, detects text regions, and runs OCR to return a compact JSON with text content, saving tokens compared to full screenshots.
