(2.4) Backports from master branch (#7854)

Original commit is a5f19f7dd6ae5bc93f73e0417a9e0bfd34c01672

All of these: (performance) Prefer prefix ++/-- operators for non-primitive types.
[modules/calib3d/src/fundam.cpp:1049] -> [modules/calib3d/src/fundam.cpp:1049]: (style) Same expression on both sides of '&&'.

The INSTALL_NAME_DIR property of a target specifies how a dynamic library should
be found on OS X.  If INSTALL_NAME_DIR is not specified the loader will search
relative to the standard search paths.  If specified it should either be
an absolute path or relative path prefixed with either @executable_path,
@load_path, or @rpath.  Specifying "lib" does not make sense here and
causes linking error as documented here:

http://answers.opencv.org/question/4134/cmake-install_name_tool-absolute-path-for-library-on-mac-osx/

and here

http://stackoverflow.com/questions/26978806/dyld-library-not-loaded-lib-libopencv-core-3-0-dylib-reason-image-not-found

This patch removes INSTALL_NAME_DIR everywhere it is set to "lib".
An alternate solution would be to set an absolute path like
"${CMAKE_INSTALL_PREFIX}/lib" or relative path like
"@executable_path/../lib".  However, if there is not specific need for
specifying a path, it is probably best left unset.

In case fillPassedSamples find zero passed negatives and leafFA is achieved, function updateTrainingSet will return false and print misleading information, this PR fix the case.

major time consuming part for training app is in function fillPassedSamples for negatives, 
this change make SetImage quicker, in a test of my own data, the total time for SetImage decrease from 9177666 to 5839263,
only help for Haar feature and non Haar::ALL mode which is the most commonly case

minor fix since predict() returun int

minor fix since predict() returun int