* remove raw NEON/SSE2 implementation as much as possible
  * replace them to universal intrinsic in InRange/Compare/AddWeighted

OpenCV fails to detect tbb on a debug build if the platform has only installed
debug libraries.  This PR adds an additional check to the tbb detect logic for
systems that only install tbb debug and not both tbb debug and release.

imgproc: use universal intrinsic as much as possible (#9714)

* use universal intrinsic as much as possible
  * make SSE3 part as common as possible with universal intrinsic implementation
  * put the reducing part out of the main loop

* follow the comment
  * fix the typo
  * use v_reduce_sum4

* follow the comment again
  * remove all CV_SSE3 part from smooth.cpp

doc: fix documentation generation with JavaScript bindings

The non-maximum suppression in the Hough accumulator incorrectly ignores maxima that extend over more than one cell, i.e. two neighboring cells both have the same accumulator value. This maximum is dropped completely instead of picking at least one of the entries. This frequently results in obvious circles being missed.

The behavior is now changed to be the same as for hough_lines.

See also https://github.com/opencv/opencv/issues/4440