Verifiable and Provably Secure Machine Unlearning