* example using the Canny algorithm with custom image gradient.

* Modified example to integrate new function canny with custom gradient

      * accumulate, accumulateSquare, accumulateProduct and accumulateWeighted

Exp and Log IPP implementations are changed to hal interface;

consequences of changes

accuracy test

Solve issue 7063

  * The difference becomes too large when multiply is done in int16
  * To reproduce the test failure, IPP has to be switched off

Program flow is not easily apparent so hopefully ANY comments in the code could help new users.

"declare.in(img, WARMUP_RNG)" overrides input image with random values

Also added support for different input sizes (via cv::resize)

cudev (only for use in .cu files) & cudalegacy (only legacy code) not included