MPL testing: the last failing test on the buildbot and future ideas

Hi,

I'm hoping to elicit the help of a developer who can fix the last
remaining failing test on the Mac buildslave. In a nutshell, there looks
to be a regression with the interaction between Axes.autoscale_view()
and the xunits kwarg of Axes.plot(). I have posted a bug report at
https://sourceforge.net/tracker/?func=detail&aid=2846058&group_id=80706&atid=560720

I'd like to ask for a bit of help fixing this one, because I'd then like
to work on a few more things, which I think will be easier (mentally, at
least), if we have all tests passing on at least one platform.

* borrow (and maybe extend) scipy's known failing test decorator so that
known issues don't elicit test failures. (E.g. font placement
differences between Ubuntu and Mac OS X. see
https://sourceforge.net/tracker/?func=detail&aid=2843243&group_id=80706&atid=560720
for more info.)

* automatically send notifications to somewhere (to where?) when tests
start failing

* write documentation for developers and patch submitters describing how
to write unit tests for MPL -- both image based and more traditional kinds.

* get my Windows virtual machine doing this same dance

* automatically upload daily builds of binaries for Win and Mac

* adding more machines to the buildbot system

Of course, I'm happy to help if anyone wants to contribute to these or
other ideas. I plan to slowly chip away at this stuff over the coming
weeks and months.

-Andrew