cv::cuda::Stream synchronization on gaussian filter

There is an internal _buf which belongs to the filter

As a result you need to use a different filter per stream. e.g.

auto gaussFilterS1 = cv::cuda::createGaussianFilter(CV_8UC1, CV_8UC1, kernelSizeGaussian, 0.0);
auto gaussFilterS1 = cv::cuda::createGaussianFilter(CV_8UC1, CV_8UC1, kernelSizeGaussian, 0.0);

This needs to be documented.