Numba Optimized Implementations for scikit bio Performance Critical Functions

scikit-bio relies on Cython extensions for performance-critical distance and ordination operations, but these cannot accept GPU arrays. This project replaces Cython functions with Numba implementations that match or exceed Cython on CPU via njit parallel and provide GPU execution via cuda jit. The work covers center_distance_matrix, Mantel permutations, PERMANOVA, geomedian, and ordination utilities. I have two open PRs (2429, 2436) adding Array API support, a working Numba prototype benchmarked against Cython, and mentor endorsement to center the proposal around this optimization approach.