RT-Sketch: Goal-Conditioned Imitation Learning from Hand-Drawn Sketches