Algebraic Sphere gradient is normalized

[ThibaultLejemble]

I think this is an error

PONCA_MULTIARCH inline VectorType primitiveGradient () const { return m_ul.normalized(); }

https://github.com/poncateam/ponca/blob/f8e54a568fe54b9ed78eb612b9ec97101c79df80/Ponca/src/Fitting/algebraicSphere.h#L252C8-L252C8

The gradient of a primitive should be different than the normal vector and should not be normalized?

But without the normalization, the test gls_paraboloid_der is broken.
If the gradient should not be normalized, then I'll check what is the issue with this test.

[nmellado]

True, there is an issue here: primitiveGradient was initially designed as being an accessor to the local surface normal (thus the normalization). But the name tells something else, and as you said, there is no reason to have the scalar field normalized.

My suggestion is to clarify the API and separate normal and gradient. This is somehow related to the local frame considered here #79.