BigQuery: Add load_table_from_json() method to BQ client

[plamut]

Closes #4553.

This PR adds a helper client method load_table_from_json().

How to test

Check that the method is similar to e.g. client.load_table_from_dataframe() and in line with the requirements.

P.S.: I will address the last issue comment separately once this is approved (replacing the _add_rows() helper in system tests is a separate refactoring task IMO).

[tswast]

Yay!

A few nitpicky things.

[tswast]

I'm having second thoughts about modifying the input config object. See:

google-cloud-python/bigquery/google/cloud/bigquery/client.py

Lines 1523 to 1529 in fcf99ce

	if job_config is None:
	job_config = job.LoadJobConfig()
	else:
	# Make a copy so that the job config isn't modified in-place.
	job_config_properties = copy.deepcopy(job_config._properties)
	job_config = job.LoadJobConfig()
	job_config._properties = job_config_properties

We might want to refactor that code into a helper function for creating or copying a LoadJobConfig.

[tswast]

Also, if there isn't a schema set, we should enable schema autodetect for them.

[plamut]

Is there a reason for not simply deep-copying the entire LoadJobConfig instance instead of separately deep-copying the _properties attribute?

The only other private attribute seems to be the _job_type string, and __init__() (that would not get called) does nothing apart from setting some attributes through setters, which more or less manipulate the same _properties dict behind the scenes.

[tswast]

One potential reason is that _properties can be used to send values to the API for Alpha releases that aren't yet supported in client libraries. Not sure if deep copy of LoadJobConfig would carry those over.

_properties is the real core of the object, since that's what defines what we end up actually including in the API request.

[plamut]

Pushed new commits, but GitHub does not detect them. Also tried rebasing the branch, still no luck. Maybe closing and re-opening will work.

Update: The same happened on the replacement PR, but then the new changes to the latter showed up after a couple of minutes. Seems like a temporary GitHub service glitch.