Supports delay-loading for Ninja generators and Python bindings.

Enable cudacodec libs by default, prevent their locations being lost when cmake is re-run and add log info when the libs/headers cannot be found.

Adds the option to enable delay loading of CUDA DLLs on Windows. This is particularly useful to use the same binary on systems with and without CUDA support without distributing the CUDA DLLs to systems that cannot use them at all due to missing CUDA-supported hardware.
Resolves #13509

warning #1394-D: field of class type without a DLL interface used in a class with a DLL interface
from windows CUDA builds.

Update CMake rules to include Nvidia Video Codec SDK encoding libs

There was a minor typo in the FATAL error message when the specified CUDA generation does not match any known generation

     - CC for RTX3090, RTX3080 and RTX3070

  * keep it possible to build for old CC
  * make sure old generations don't come up for the choice
  * remove related version check of old one

     * treat both CMAKE_C_COMPILER and c_compiler_realpath as candidate

- preventive fix for arch "11.0" (CUDA_ARCH_BIN_OR_PTX_10 bug)
- new var: OPENCV_CUDA_DETECTION_NVCC_FLAGS
- new var: OPENCV_CMAKE_CUDA_DEBUG

* core(cuda): fix build

- MSVS 19.25.28612.0
- CUDA release 11.0, V11.0.167

* cmake(cuda): backport workaround for CUDA 11

* cmake(cuda): call CUDA_BUILD_CLEAN_TARGET() on finalize

* cmake(cuda): use CMAKE_SUPPRESS_REGENERATION with MSVS

Fix cuda11

* use cudnn_version.h to detect version when it is available

* remove nppi from CUDA11

* use ocv_list_filterout

* dnn(cuda): temporary disable CUDNN 8.0

  * use only supported CC in the list
  * workaround of #17526

  * Ampere has CC 8.0
  * Arm64 server support has been added in CUDA 11 (only V100 for now)

  * detect header automatically and not based on version number

  * cache the result
  * DRY
  * brush up based on review

* cmake: avoid deprecated LINK_PRIVATE/LINK_PUBLIC

see CMP0023 (CMake 2.8.12+)

* cmake: fix 3rdparty list

- don't include OpenCV modules

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

CMake 3.9.0+ is required

issue 11552