keep cufftPlan2d across ConvolveImpl::convolve calls

[r2d3]

on some CUDA versions creating/destroying cufftPlan2d is very time consuming we now create/destroy them in ConvolveImpl::create() when dft_size change and destroy them in the dtor

this solves issue #3385

Pull Request Readiness Checklist

See details at https://github.com/opencv/opencv/wiki/How_to_contribute#making-a-good-pull-request

• I agree to contribute to the project under Apache 2 License.
• To the best of my knowledge, the proposed patch is not based on a code under GPL or another license that is incompatible with OpenCV
• The PR is proposed to the proper branch
• There is a reference to the original bug report and related work
• There is accuracy test, performance test and test data in opencv_extra repository, if applicable
  Patch to opencv_extra has the same branch name.
• The feature is well documented and sample code can be built with the project CMake

[asmorkalov]

👍 Thanks a lot for the contribution. Tested manually with Ubuntu 18.04, CUDA 10.2, GF 1080. Performance improvement is visible even with existing OpenCV perf tests:

Geometric mean (ms)

                   Name of Test                     default after   after   
                                                            cufft   cufft   
                                                             fix     fix    
                                                                      vs    
                                                                   default  
                                                                  (x-factor)
Convolve::Sz_KernelSz_Ccorr::(1280x720, 17, false)   1.242  0.558    2.23   
Convolve::Sz_KernelSz_Ccorr::(1280x720, 17, true)    1.233  0.567    2.18   
Convolve::Sz_KernelSz_Ccorr::(1280x720, 27, false)   1.200  0.641    1.87   
Convolve::Sz_KernelSz_Ccorr::(1280x720, 27, true)    1.206  0.523    2.30   
Convolve::Sz_KernelSz_Ccorr::(1280x720, 32, false)   1.197  0.527    2.27   
Convolve::Sz_KernelSz_Ccorr::(1280x720, 32, true)    1.181  0.600    1.97   
Convolve::Sz_KernelSz_Ccorr::(1280x720, 64, false)   1.178  0.524    2.25   
Convolve::Sz_KernelSz_Ccorr::(1280x720, 64, true)    1.197  0.524    2.28   
Convolve::Sz_KernelSz_Ccorr::(1280x1024, 17, false)  1.423  0.816    1.74   
Convolve::Sz_KernelSz_Ccorr::(1280x1024, 17, true)   1.419  0.813    1.74   
Convolve::Sz_KernelSz_Ccorr::(1280x1024, 27, false)  1.428  0.825    1.73   
Convolve::Sz_KernelSz_Ccorr::(1280x1024, 27, true)   1.428  0.825    1.73   
Convolve::Sz_KernelSz_Ccorr::(1280x1024, 32, false)  1.419  0.820    1.73   
Convolve::Sz_KernelSz_Ccorr::(1280x1024, 32, true)   1.439  0.821    1.75   
Convolve::Sz_KernelSz_Ccorr::(1280x1024, 64, false)  1.416  0.829    1.71   
Convolve::Sz_KernelSz_Ccorr::(1280x1024, 64, true)   1.433  0.821    1.74   
Convolve::Sz_KernelSz_Ccorr::(1920x1080, 17, false)  1.978  1.278    1.55   
Convolve::Sz_KernelSz_Ccorr::(1920x1080, 17, true)   1.972  1.275    1.55   
Convolve::Sz_KernelSz_Ccorr::(1920x1080, 27, false)  1.993  1.289    1.55   
Convolve::Sz_KernelSz_Ccorr::(1920x1080, 27, true)   1.980  1.281    1.55   
Convolve::Sz_KernelSz_Ccorr::(1920x1080, 32, false)  1.979  1.280    1.55   
Convolve::Sz_KernelSz_Ccorr::(1920x1080, 32, true)   1.978  1.276    1.55   
Convolve::Sz_KernelSz_Ccorr::(1920x1080, 64, false)  1.985  1.281    1.55   
Convolve::Sz_KernelSz_Ccorr::(1920x1080, 64, true)   1.984  1.278    1.55