ENH: Add doctest_skip_parser allowing conditional skipping of doctests

[JDWarner]

This framework is based on a DiPy pull request: dipy/dipy#105

Our first conditional doctests are needed if neither PyQt nor PySide were available in skimage.viewer, reported by @cdeil. However, both instances were actually class docstrings rather than function docstrings, which was not supported by the DiPy decorator.

Here I have generalized the framework slightly (along with updated tests for both classes and functions) and it appears now to work transparently for either case.

I believe this resolves the problem noted after the merge of #811, but @cdeil should test this to make sure.

@ahojnnes, @stefanv and other devs might want to check over current and future skipped doctests; some can probably be converted into conditionals rather than outright skips.

[JDWarner]

I believe Travis is hanging because we try to fool it into thinking it has some form of Qt when it does not.

Thus the qt_api is incorrectly reported and these tests are run instead of skipped. This won't happen on any production system (test it for yourself).

Perhaps we need to fool Travis in a different way?

[ahojnnes]

@JDWarner Hm... Qt is actually installed and working on Travis, but I think the show command freezes until the window is closed?

[JDWarner]

@ahojnnes of course...

Should work now.

[stefanv]

Can't a person have a condition for the entire test?

[JDWarner]

I believe doctests just don't support skipping blocks. There was some discussion about this around #811.

[coveralls]

Coverage remained the same when pulling beb395a on JDWarner:add_conditional_doctest_decorator into f8b5550 on scikit-image:master.

[ahojnnes]

Perfect, thanks Josh!

[ahojnnes]

@stefanv skipping entire blocks is not possible and the python devs decided that this feature is not wanted: http://bugs.python.org/issue9017

[JDWarner]

@cdeil Can you confirm this resolves the issue lingering after #811?