Add ocl version BackgroundSubtractorKNN (#10553)

* Add ocl version bgfg_knn

* Add ocl KNN perf test

* ocl KNN: Avoid unnecessary initializing when non-UMat parameters are used

* video: turn off OpenCL for color KNN on Intel devices

due performance degradation

* video: turn off KNN OpenCL on Apple devices with Intel iGPU

due process freeze during clBuildProgram() call

prepareData() bug feeds the same image (the latest)
OCL perf test doesn't pass accuracy(!) check now, so this check is disabled.

  * confirmed test failure on Jetson TX1 and TX2
  * show the performance but not bit exact result

during pre-commit testing

general:
- all iterative tests have been replaced with parameterized tests
- old-style try..catch tests have been modified to use EXPECT_/ASSERT_ gtest macros
- added temporary files cleanup
- modified MatComparator error message formatting

imgcodecs:
- test_grfmt.cpp split to test_jpg.cpp, test_png.cpp, test_tiff.cpp, etc.

videoio:
- added public HAVE_VIDEO_INPUT, HAVE_VIDEO_OUTPUT definitions to cvconfig.h
- built-in MotionJPEG codec could not be tested on some platforms (read_write test was disabled if ffmpeg is off, encoding/decoding was handled by ffmpeg otherwise).
- image-related tests moved to imgcodecs (Videoio_Image)
- several property get/set tests have been combined into one
- added MotionJPEG test video to opencv_extra

Tests of the mask are also included.

This is useful for registering a non-square image against a non-square
template.

This also needs to relax a sanity check as per
https://github.com/Itseez/opencv/pull/3851

The matrix templateZM needs to be initialized because otherwise
uninitialized values leak into the correlation in:

    const double correlation = templateZM.dot(imageWarped)

In the worst case this will lead the correlation to be NaN ruining the
whole routine. The subtraction does not initialize templateZM due to the
mask.

Unfortunately, the uninitialized values (by altering the correlation)
have the side effect of dragging out the computation a little longer
giving a slightly better error bound. This means that fixing this bug
breaks perf_ecc where

    SANITY_CHECK(warpMat, 1e-3);

is just a little too tight and happens to work due to the uninitialized
values. Since this is a performance not a accuracy test I think it is OK
to just relax the error bound a little bit (the tight error bound being
after all the result of a bug).

Looks like we don't actually use them; when creating precompiled headers
with Visual C++, a dummy .cpp is created at build time.

Also cv::, cv::gpu:: and cv::ocl:: namespace prefixes can be safely omitted
inside CV_ENUM and CV_FLAGS

e.g. <opencv2/core/core.hpp> become <opencv2/core.hpp>

Also renamed <opencv2/core/opengl_interop.hpp> to <opencv2/core/opengl.hpp>

Evangelidis, G.D. and Psarakis E.Z. "Parametric Image Alignment using Enhanced
Correlation Coefficient Maximization", IEEE Transactions on PAMI, vol. 32, no.
10, 2008

* added DenseOpticalFlow interface
* moved OpticalFlowDual_TVL1 to src folder