Introduce GpuMatT for ALL [Input]OutputArrays

[shrubb]

Currently:

struct TensorWrapper gammaCorrectionCuda(struct cutorchInfo info,
        struct TensorWrapper src, struct TensorWrapper dst, bool forward)
{
    cuda::GpuMat retval = dst.toGpuMat();
    cuda::gammaCorrection(src.toGpuMat(), retval, forward, prepareStream(info));
    return TensorWrapper(retval, info.state);
}

Need (everywhere!):

struct TensorWrapper gammaCorrectionCuda(struct cutorchInfo info,
        struct TensorWrapper src, struct TensorWrapper dst, bool forward)
{
    GpuMatT retval = dst.toGpuMatT();
    cuda::gammaCorrection(src.toGpuMat(), retval, forward, prepareStream(info));
    return TensorWrapper(retval, info.state);
}