[tune] Cannot optimize a metric nested in result

[ghost]

What is the problem?

The script at the bottom failed because it pass evaluation/episode_reward_mean as the metric to optimize. In the reported dict result, episode_reward_mean is nested under evaluation.
After digging a bit into the source code, I think this kind of metric could be optimize in tune. But, the validation process failed because it use nested result.

Ray version and other system information (Python version, TensorFlow version, OS):
ray: 2.0.0

Reproduction (REQUIRED)

Please provide a short code snippet (less than 50 lines if possible) that can be copy-pasted to reproduce the issue. The snippet should have no external library dependencies (i.e., use fake or mock data / environments):

import ray
from ray.rllib.agents.dqn import dqn
from ray import tune


if __name__ == "__main__":
    ray.init(local_mode=True, num_cpus=2)

    config = {
        "env": "CartPole-v1",
        "framework": "torch",

        "timesteps_per_iteration": 10,
        "evaluation_interval": 1,
        "evaluation_num_episodes": 1,
    }

    analysis = tune.run(
        dqn.DQNTrainer,
        config=config,
        metric="evaluation/episode_reward_mean",
        mode="max",
        num_samples=10,
        stop={
            "evaluation/episode_reward_mean": 20
        }
    )

    ray.shutdown()

If the code snippet cannot be run by itself, the issue will be closed with "needs-repro-script".

• I have verified my script runs in a clean environment and reproduces the issue.
• I have verified the issue also occurs with the latest wheels.