Add educational UMAP implementation to machine_learning

### Feature description ## Description This issue proposes adding an educational implementation of Uniform Manifold Approximation and Projection (UMAP) to the `machine_learning` section. Features: - pairwise distance computation - nearest-neighbour graph construction - simplified fuzzy membership strengths - low-dimensional embedding optimisation - doctests and type hints - Iris dataset example The implementation is simplified for educational purposes and follows the repository style used in dimensionality reduction algorithms such as PCA and t-SNE.