RecognizeTextTool
Extracts text from a local image with on-device ML Kit text recognition (Latin script). Returns the full recognized text plus per-line text and bounding boxes.
No permissions required; image_path is sandboxed to the external-storage root via PathValidator and must point at an existing file. Read-only.
Output keys: text, line_count, lines (each with text and nullable bounding_box {left, top, right, bottom}).