misplaced CV_WRAP statement lead to not exposing EM::getCovs() to python/java/matlab wrappers

add a CV_OUT to return the covs properly

Now StereoSGBM behaves similarly to StereoBM in terms of ROI handling.

New mode is approximately 2-3 times faster than MODE_SGBM
with minimal degradation in quality and uses universal
HAL intrinsics. A performance test was added. The accuracy
test was updated to support the new mode.

Signed-off-by: Yan Wang <yan.wang@linux.intel.com>

It could reduce 7 multiplications every iteration.
Signed-off-by: Yan Wang <yan.wang@linux.intel.com>

rewrite & change convertFromGLBuffer() & convertToGLBuffer() into acquireGLBuffer() & releaseGLBuffer(), respectively

opengl sample: added buffer support

tested and fixed buffer support on Windows

change glFlush() call to glFinish()

added UMat::release() call; fixed functions' names

adopted & implemented API suggestion(s) from Alexander

fixed unreachable code warning

added more info to the mapGLBuffer/unmapGLBuffer description

changed output from fps to time (in msec). Notes for dx10 and dx11: the first measure is wrong (zero), the followiing are correct; measured time includes output text to surface.

pos should never exceed strlen, else the overflow check 'if (!len)' will wrongfully pass

1. Remove uncessary index calculation.
2. Use mad/mad24 as possible.
Signed-off-by: Yan Wang <yan.wang@linux.intel.com>

by this new function we can set libjpeg "scale_denom" parameter and load jpeg images scaled 1/2 - 1/4 - 1/8
other image formats resized after loading