QoS-Aware Power Minimization of Distributed Many-Core Servers using Transfer Q-Learning