Trim pydata-google-auth package and add tests

[tswast]

This is the initial version of the proposed pydata-google-auth package (to be used by pandas-gbq and ibis). It includes two methods:

• pydata_google_auth.default()
  • A function that does the same as pandas-gbq does auth currently. Tries google.auth.default() and then falls back to user credentials.
• pydata_google_auth.get_user_credentials()
  • A public get_user_credentials() function, as proposed in pandas-gbq auth proposal googleapis/python-bigquery-pandas#161. Missing in this implementation is a more configurable way to adjust credentials caching. I currently use the reauth logic from pandas-gbq.

I drop try_credentials(), as it makes less sense when this module might be used for other APIs besides BigQuery. Plus there were problems with try_credentials() even for pandas-gbq (googleapis/python-bigquery-pandas#202, googleapis/python-bigquery-pandas#198).

FYI @max-sixty @cpcloud