Learning to Accelerate Partitioning Algorithms for the Global Optimization of Nonconvex Quadratically-Constrained Quadratic Programs