As TensorSpace can auto detect model's input ( #165 ), user do not need to provide all input shapes, to make loader API more friendly, deprecate multiInputs and inputShapes in Loader.