Learning Rigidity-Based Flocking Control Using Gaussian Processes with Probabilistic Stability Guarantees