[ENH] Add CredentialsCache classes

[tswast]

The CredentialsCache classes allow for more flexible caching of user
credentials to disk. The pydata_google_auth.cache module provides 3
canonical instances of caches:

• pydata_google_auth.cache.READ_WRITE : (default) read and write to a file in ~/.config/..., just like pandas-gbq does by default,
• pydata_google_auth.cache.REAUTH : write to disk but don't read from it, just like pandas-gbq with reauth=True,
• pydata_google_auth.cache.NOOP : don't read/write to disk at all. Useful when running on shared machines.

I believe this is the final implementation detail needed before releasing 0.1.0 of pydata-google-auth. The public interface includes:

• pydata_google_auth.default(scopes, client_id=(pydata-auth project), client_secret=(pydata-auth project), credentials_cache=READ_WRITE, auth_local_webserver=False) : Same as current pandas-gbq behavior. Does google.auth.default() with fallback to user authentication.
• pydata_google_auth.get_user_credentials(scopes, client_id=(pydata-auth project), client_secret=(pydata-auth project), credentials_cache=READ_WRITE, auth_local_webserver=False) : Get user credentials (loaded from cache, if available) via the 3-legged-oauth flow.

I'll update the docs to remove leftovers from the pandas-gbq fork in a subsequent PR.

Towards pandas-gbq auth rewrite: googleapis/python-bigquery-pandas#161

Closes #1

@craigcitro please review

[tswast]

I haven't fully configured Travis yet. Tests pass locally.

$ pytest tests
============================== test session starts ===============================
platform darwin -- Python 3.6.4, pytest-3.7.1, py-1.5.4, pluggy-0.7.1
rootdir: /Users/swast/src/pydata-google-auth, inifile:
collected 7 items

tests/system/test_auth.py .....                                            [ 71%]
tests/unit/test_auth.py ..                                                 [100%]

=========================== 7 passed in 12.66 seconds ============================

[max-sixty]

When do you think we should move pandas-gbq to be dependent on this?

[tswast]

I need to make a release of this package first. :-) I think it's "code complete", but I'm working on getting docs to build, tests running, etc.

[max-sixty]

OK great! This is going to be v useful, and I'm happy it can spread wider than pandas-gbq