CUDA implementation wants to convert std::vector<KeyPoint> <-> GpuMat.
There is no direct mapping from KeyPoint (mix of int/float fields)
into cv::Mat element type, so this conversion must be avoided.

Legacy mode is turned back for CUDA builds.

CMake: Building Dynamic Framework on iOS (#8009)

* Updated python script with dynamic parameter

Updated python script to build static library by default

Updated python script to include bitcode flag

Added bitcode flag to c flags

Fixed directories and targets with static

Bitcode parameter fixed

Fixed script for static library

Fixed parameters in build function

Updated cmake common toolchain

Added changes to OpenCV Utils

Updates to cmake

Added cache internal

Updates to common toolchain

Fixed path in framework destination and added UIKit dependency

Dynamic plist for framework

Lib version removed hardcoded value

Removed trailing whitespace in toolchain

* Removed trailing whitespace

* Fixed typo in comment

* Renamed bitcode variable to bitcodedisabled

* Fixed target device family

"install DIRECTORY does not allow "COMPONENT" after PATTERN or REGEX."

Unfortunately there is no stable support for installation of PDB files in CMake.
This patch is just eliminate problems with modern versions.

(2.4) Backports from master branch (#7854)

Useful for debug purposes:
    cmake -DDEBUG_POST="ocv_cmake_dump_vars(CUDA)" .
    cmake -DCMAKE_PRE="ocv_cmake_dump_vars(\"OPENCL|opencl\")" .

1) fix include directories
2) don't use PCH for source file with custom COMPILE_FLAGS

Also put "object" targets to off by default (they doesn't work well)

CMake doesn't work with empty paths ("") or paths which contains "./"