[tune] logging fixes, better warnings, better cluster support

[richardliaw]

Robust CSV writing, etc.

Fixes #3834, fixes an issue raised in #3904, fixes an issue raised in #3891.

[AmplabJenkins]

Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-PRB/11313/
Test PASSed.

[AmplabJenkins]

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-PRB/11339/
Test FAILed.

[AmplabJenkins]

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-PRB/11336/
Test FAILed.

[AmplabJenkins]

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-PRB/11340/
Test FAILed.

[AmplabJenkins]

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-PRB/11342/
Test FAILed.

[richardliaw]

compare without done=True -

import ray
from ray.tune import run_experiments


def train_model(config, reporter):
    # compare without done=True
    reporter(mean_accuracy=1, done=True)


if __name__ == '__main__':
    ray.init()
    run_experiments({"my_exp": {"run": train_model}})

[AmplabJenkins]

Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-PRB/11343/
Test PASSed.

[AmplabJenkins]

Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-PRB/11434/
Test PASSed.