A fast novel heuristic and machine learning-assisted hybrid metaheurics for the Euclidean leaf-constrained minimum spanning tree problem