add external_udf_uris support to query()

[robin900]

...with docstring and test coverage. Change write_to_table code to use the same
list comprehension to construct userDefinedFunctionResources.

[tylertreat]

lgtm, just one thing regarding the kwarg. Thanks!

[tylertreat]

We shouldn't use lists as a default value for kwargs (see http://reinout.vanrees.org/weblog/2012/04/18/default-parameters.html). This should default to None and check if it's None and assign it to an empty list.

[robin900]

So I've fixed the existing write_to_table's default arg as well, in a force update.

[robin900]

Are you up to making a quick 1.9.1 release to PyPI with this change? I'm using BigQuery-Python in my production code, and would like to avoid sourcing this package from my own git repo.

[tylertreat]

@robin900 I will try to get a release made later today.

[robin900]

@tylertreat So the bad news is that query() -- that is, the synchronous query endpoint in the v2 REST API (POST to /queries) -- doesn't support external UDFs, and there are no plans to add it. Instead, only asynchronous queries -- a POST to the /jobs endpoint -- allow support for UDFs.

I see two possible solutions:

1. Implement BigQueryClient.query so that it uses self._insert_job(query_job_request), and does a wait_on_job and then returning when job is complete.
2. Doing Move import to avoid possible ImportError. #1 but only in cases where external_udf_uris are present.

What are your thoughts?

An additional need: external UDFs can be inlined via inlineCode instead of resourceUri. I have a local change that accepts both external_udf_uris and external_udf_inlines as arguments to query(). Shall I include that additional argument for all relevant client methods?