Improve string representations of entities

[smurching]

Currently, MLflow entities use default Python string serialization, resulting in hard-to-interpret output like:

>>> from  mlflow.tracking import get_service
>>> service = get_service()
>>> experiments = service.list_experiments()
[<mlflow.entities.experiment.Experiment at 0x7f93a8902a90>]

This PR improves entities' string representations to include a listing of their fields (and recursive serialization of lists of entities, e.g. lists of metrics/params within a run). The above now results in:

>>> from  mlflow.tracking import get_service
>>> service = get_service()
>>> experiments = service.list_experiments()
[<mlflow.entities.experiment.Experiment: 'artifact_location'='/Users/sid/code/mlflow/mlruns/0', 'experiment_id'=0, 'name'='Default'>]

Runs now appear as:

>>> r = service.get_run(run_id)
>>> r
<mlflow.entities.run.Run: 'data'=<mlflow.entities.run_data.RunData: 'metrics'=[<mlflow.entities.metric.Metric: 'key'='random_int', 'timestamp'=1534987517, 'value'=75.0>, <mlflow.entities.metric.Metric: 'key'='foo', 'timestamp'=1534987517, 'value'=7.0>], 'params'=[<mlflow.entities.param.Param: 'key'='param1', 'value'='5'>], 'tags'=[]>, 'info'=<mlflow.entities.run_info.RunInfo: 'artifact_uri'='/Users/sid/code/mlflow/mlruns/0/145d84e1b29c4fe9a3b191f27e63a79f/artifacts', 'end_time'=1534987517756, 'entry_point_name'='', 'experiment_id'=0, 'name'='Run 6', 'run_uuid'='145d84e1b29c4fe9a3b191f27e63a79f', 'source_name'='example/advanced/test_remote_server.py', 'source_type'=4, 'source_version'='e375c52679ab74b3f6e864d6bcab0118125e31e4', 'start_time'=1534987517633, 'status'=3, 'user_id'='sid'>>

Feedback is welcome - run serialization is still somewhat unwieldy due to the number of fields in a run.

[mparkhe]

Very cool implementation. Minor comment about property ordering in output.

[mparkhe]

Can we iterate this in the order specified _properties() method for each entity.

Here's why with an example. Experiment would read more intuitively this way --

[<mlflow.entities.experiment.Experiment: 'experiment_id'=0, 'name'='Default', 'artifact_location'='/Users/sid/code/mlflow/mlruns/0'>]

Rather than hash sorted order.

[mateiz]

+1, that would be nice. Also don't put quotes around the key -- basically change the self.serialize(key) to just key.

[mparkhe]

👍 very neat and tight implementation!

[mateiz]

Just as a note, I think "serialize" can mean converting to formats other than string representation, so I would call this to_string or pretty_print or something.

[mateiz]

I'd include just the last part of the class name: Experiment instead of mlflow.entities.experiment.Experiment.

[smurching]

Thanks @mateiz @mparkhe, merging to master!