EFKAN: A KAN-Integrated Neural Operator For Efficient Magnetotelluric Forward Modeling