Change storage ordering of the eigen vectors descriptors

[aguenette]

The eigen vectors were serialized row major instead of column major, which shouldn't be the case since everything else is ordered column major in libpm. So, it should be assumed that it's also the case for the eigen vectors. This PR will have an impact the users using the following filters:

• Ellipsoids (1), Ellipsoids (2)
• Gestalt (1), Gestalt (2)
• SamplingSurfaceNormal
• SurfaceNormal

and the following descriptors:

• keepEigenVectors
• keepCovariances

So, retrieving the eigen vectors from a 3D point cloud before this PR could be done like this:

PM::Matrix eigenVectors = cloud.getDescriptorsViewByName("eigVectors"); // [ a₁, b₁, c₁, a₂, b₂, c₂, a₃, b₃, c₃ ]ᵀ    

Eigen::Map<PM::Vector, 0, Eigen::InnerStride<3>> a(eigenVectors.col(n).data(), 3);     // [ a₁, a₂, a₃ ]ᵀ
Eigen::Map<PM::Vector, 0, Eigen::InnerStride<3>> b(eigenVectors.col(n).data() + 1, 3); // [ b₁, b₂, b₃ ]ᵀ
Eigen::Map<PM::Vector, 0, Eigen::InnerStride<3>> c(eigenVectors.col(n).data() + 2, 3); // [ c₁, c₂, c₃ ]ᵀ

And after this PR it could be done like this:

PM::Matrix eigenVectors = cloud.getDescriptorsViewByName("eigVectors"); // [ a₁, a₂, a₃, b₁, b₂, b₃, c₁, c₂, c₃ ]ᵀ    

PM::Vector a = eigenVectors.block(0, n, 3, 1); // [ a₁, a₂, a₃ ]ᵀ
PM::Vector b = eigenVectors.block(3, n, 3, 1); // [ b₁, b₂, b₃ ]ᵀ
PM::Vector c = eigenVectors.block(6, n, 3, 1); // [ c₁, c₂, c₃ ]ᵀ

Then, reconstructing the matrix is the same for both method:

PM::Matrix R(3,3);
R << a, b, c;

As we can see, it is more straightforward to retrieve the eigen vectors when they ordered column major.

[pomerlef]

@aguenette update to current master.

[aguenette]

@pomerlef It's merge.