On Maximal Rewards and $|varepsilon$-Optimal Policies in Continuous Time Markov Decision Chains