refactor: DQN docstring & fix divergence

[kkweon]

1. Make it as close as implemented in the paper
2. Fix slow iteration (tf.Operation inside a loop is too slow)
3. Clean up codes & remove api_key
   • hunkim is spamming in https://gym.openai.com/envs/CartPole-v0

Result

2013 ver Game Cleared within 287 episodes with avg reward 199.18

2015 ver Game Cleared in 265 episodes with avg reward 199.39