Add gradient magnitude computation program LN2_GRAMAG

[ofgulban]

It is currently becoming more and more handy to have a fast / ram efficient gradient magnitude computation. I should translate the gradient magnitude methods from https://github.com/ofgulban/segmentator to LayNii.

• Add gradient magnitude without RAM inflation (do not hold gradient images, directly compute gradient magnitude per voxel). Important for e.g. whole brain 0.1 mm human brain images.
• Add circular difference option for phase images (-pi to pi range).

[ofgulban]

I should add a circular difference option to compute gradients on phase images (e.g. -pi to pi).

[ofgulban]

Circular difference option added

See the images below for the difference between normal gradient magnitude and "circular" gradient magnitude outputs:

• Upper left is a phase image.
• Upper right is "normal" gradient magnitude.
• Lower left if circular gradient magnitude.
• Colormaps are equal between gradient magnitude images.
• Incorrectly inflated differences around phase wraps are visible in the "normal" gradient magnitude image (upper right).

Simulated phase data results:

Empirical phase data results: