V0.2.0

[rjanvier]

• Improve 'selected' method to be closer to Jakteristics: no preallocation of the whole result set. The result set it is grown when needed. It results in a less efficient method but it should be even more memory friendly.
• Optimize 'selected' method: use a dynamic scheduler, sort NN by distance when needed by a partial_sort.
• PCA computation use Eigen::SelfAdjointEigenSolver, since the covar is symetric. It is slighty faster (complexity is $9n^3$ instead of $25n^3$ where n is the size of the matrix). We now use transpose instead of adjoint (adjoint is for complex matrix) in the covariance computation.
• Misc. formatting and typo corrections

[drprojects]

Thanks @rjanvier, approved !