add cuDNN dependency and setup build for cuda4dnn (#14660)

* update cmake for cuda4dnn

- Adds FindCUDNN
- Adds new options:
   * WITH_CUDA
   * OPENCV_DNN_CUDA
- Adds CUDA4DNN preprocessor symbol for the DNN module

* FIX: append EXCLUDE_CUDA instead of overwrite

* remove cuDNN dependency for user apps

* fix unused variable warning

oss-fuzz: https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=13314
ported commit: https://github.com/glennrp/libpng/commit/a142b7a543d16d271b8aba3ef7ca4aa9a368f55d

video(DISOpticalFlow): update OpenCL implementation (#14473)

* video(DIS): add code for profiling

* video(DIS): fix test parameters

* video(DIS): simplify OpenCL kernels

- parameters -> defines
- avoid float3
- const / local scope
- improve readability, replace Kernel::set() -> args()

* video(DIS): use CV_32FC2 buffers

Merge pull request #14641 from thangktran:thangktran/bugfix/DISOpticalFlow-segfault-for-small-images

VideoCapture Direct Show. Added getter for CV_CAP_PROP_CONVERT_RGB, CAP_PROP_CHANNEL properties. (#14592)

* Added getter for CV_CAP_PROP_CONVERT_RGB, CAP_PROP_CHANNEL properties.
Some refactoring.

* One space less.

Fix dnn visualizer (#14537)

* Fixed print layer params

* Fix print

* java: fix java missing_consts

* java: test Calib3d constants

Usage example:
- run.py -a -t java --package=calib3d

HSV and HLS color conversions rewritten to wide intrinsics (#14616)

* RGB2HSV_b vectorized

* RGB2HSV_f: widen

* RGB2HSV_f: shorten, more intuitive

* HSV2RGB_f and HSV2RGB_b widen

* hls2rgb_f widen

* instrumentation instead vx_cleanup

* RGB2HLS_f widen

* RGB2HLS_b rewritten to wide universal intrinsics

* define guard against no SIMD code

* hls2rgb_b rewritten

* extra define removed

* warning fixed

* hls2rgb_b: performance fixed