Support unicode in metric data better.

[squirrelsc]

Environment:

• NNI version: master(nnictl, post v1.6)/1.4(worker)
• NNI mode (local|remote|pai): pai
• Client OS: Ubuntu
• Server OS (for remote mode only):
• Python version: py3.x
• PyTorch/TensorFlow version: PyTorch
• Is conda/virtualenv/venv used?: No
• Is running in Docker?: Yes

What issue meet, what's expected?:

note, it may be caused by an inconsistent old version of NNI, and it needs to be monitored. If there are similar issues, it can be fixed by this solution.

This metric message, which is received in nniDataStore.storeMetricData causes an error

{"sequence": 0, "trial_job_id": "Do57b", "value": "{\\\\"default\\\\": 0.0006907543685855333, \\\\"detail\\\\": [0.0009301810563630108, 0.0007786923064047202, 0.00036338974298886936], \\\\"flops\\\\": 45699600.0}", "type": "FINAL", "parameter_id": 0}

SyntaxError: Unexpected token d in JSON at position 55

It causes by str.encode('utf8') in python code is not processed correctly. And it needs to be unescaped like below.

JSON.parse('"' + str.split('"').join('\\"') + '"');

[liuzhe-lz]

Should not be a problem in latest version.