Fix canny

[rolalaro]

Fix two problems :

• in vpCannyEdgeDetection, fix the way the gradient orientation is computed, getting results that are much closer to the ones obtained using opencv-backend
• in vpImageFilter, fix a mistake on the way the partial derivatives are normalized when using opencv-backend

[fspindle]

To be complete, you could precise that this angle is between -180 and +180 deg (or \f$-\pi,+\pi\f$ see below)

[fspindle]

Is there a good reason to consider degrees instead of rad? Using deg means that to have to call vpMath::rad() that is useless
antan2() is returning rad
You can use M_PI_2 and M_PI_4 macros

[fspindle]

I suggest to rename positiveTheta into gradientOrientation to be more explicit

[rolalaro]

These suggestions have been addressed in the new commit e459437

[codecov]

Codecov Report

Attention: 6 lines in your changes are missing coverage. Please review.

Comparison is base (e49493a) 54.41% compared to head (2b4e240) 54.43%.

❗ Current head 2b4e240 differs from pull request most recent head 43f9441. Consider uploading reports for the commit 43f9441 to get more accurate results

Files	Patch %	Lines
modules/core/src/image/vpCannyEdgeDetection.cpp	0.00%	3 Missing ⚠️
modules/core/src/image/vpImageFilter.cpp	0.00%	3 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1287      +/-   ##
==========================================
+ Coverage   54.41%   54.43%   +0.01%     
==========================================
  Files         400      400              
  Lines       50512    50514       +2     
==========================================
+ Hits        27487    27496       +9     
+ Misses      23025    23018       -7     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.