normalize base64 output data in tests

[minrk]

checks for base64 the lazy way, by trying to decode it first.

avoids spurious failure when equivalent base64 data appears changed due to changes in presence of newlines, etc.

cc @takluyver this is the only failure on jack-of-none: https://travis-ci.org/jupyter/test-grid/jobs/266020642

[mpacer]

looks like there's something wrong with python-nightly and gcc, but that the python2 error actually has something to do with the way the base64 was encoded… Seemingly it lost some padding =s at its end?

[minrk]

Should be fixed now.

[mpacer]

Sounds good. I still am confused as to why both base64 and binascii exist as separate modules but this fix looks good.

[minrk]

From the docs: there are three APIs for base64 encoding in two modules.

• binascii implements efficient, low-level b2a, a2b encodings, and is meant to be primarily used via other modules
• base64 provides 'high level' API wrappers around it
• base64.encode/decodebytes is the 'legacy' API in base64, which encodes base64 as multi-line text
• base64.b64encode/decode is the 'modern' API (new in Python 2.4), and probably what we should be using in general