add tool for supervise python api

[guochaorong]

resolve #11556

[panyx0718]

LG. Do you have an example output?

[panyx0718]

perhaps we only need to parse_args once at the beginning?

[guochaorong]

yes , it will be clearer

[panyx0718]

you can put ``` at the next line.
Would be good to use the same code style as paddle-core's code.

[guochaorong]

got it!

[guochaorong]

output are as below:
{
"ParallelExecutor.init": {
"keywords": "kwargs",
"args": ["self", "use_cuda", "loss_name", "main_program", "share_vars_from", "exec_strategy", "build_strategy", "num_trainers", "trainer_id"],
"defaults": "(None, None, None, None, None, 1, 0)",
"varargs": null
},
"ParallelExecutor.bcast_params": {
"keywords": null,
"args": ["self"],
"defaults": "None",
"varargs": null
},
"ParallelExecutor.run": {
"keywords": null,
"args": ["self", "fetch_list", "feed", "feed_dict"],
"defaults": "(None, None)",
"varargs": null
}
}

if api changed, error are as below:
Exception: api changed: ParallelExecutor.init

before: {
u 'keywords': u 'kwargs',
u 'args': [u 'self', u 'use_cu**', u 'loss_name', u 'main_program', u 'share_vars_from', u 'exec_strategy', u 'build_strategy', u 'num_trainers', u 'trainer_id'],
u 'defaults': u '(None, None, None, None, None, 1, 0)',
u 'varargs': None
}
now: {
'keywords': 'kwargs',
'args': ['self', 'use_cuda', 'loss_name', 'main_program', 'share_vars_from', 'exec_strategy', 'build_strategy', 'num_trainers', 'trainer_id'],
'defaults': '(None, None, None, None, None, 1, 0)',
'varargs': None
}

[guochaorong]

oh， 想起来，于洋老师改了一版优雅版的交了~
https://github.com/PaddlePaddle/Paddle/pull/11573/files