algo: Improve Rust DR algorithm implementations with comprehensive tests

[GeorgePearse]

Summary

• Add Barnes-Hut quadtree for O(N log N) t-SNE force approximation, enabling scalability to larger datasets
• Parallelize t-SNE and Isomap with Rayon for significant speedup on multi-core systems
• Replace O(n³) Floyd-Warshall with O(n² log n) parallel Dijkstra in Isomap geodesic computation
• Add SMACOF metric MDS algorithm with stress metric for quality measurement

Changes

File	Lines	Description
src/barnes_hut.rs	+290	New Barnes-Hut quadtree implementation
src/tsne.rs	+237	Parallel distance/gradient computation
src/isomap.rs	+215	Dijkstra-based geodesic distances
src/mds.rs	+171	SMACOF algorithm + stress computation
src/pca.rs	+127	Explained variance ratio
src/lle.rs	+76	Regularized weight computation

Benchmark Results

Algorithm	Time	Trustworthiness (k=15)
t-SNE	12.69s	0.990
UMAP	7.25s	0.985
PaCMAP	0.14s	0.978
MDS	6.40s	0.889
PCA	0.00s	0.829

All algorithms include comprehensive unit tests.