Shared test infrastructure

I'm hoping the common test infrastructure makes a different for people doing cross filesystem work; it should take a lot of the grunt work out of that.

But for that to be useful we're all going to have to work on dealing with failing tests one way or the other, so that the results are useful to someone who isn't intimately familiar with a given fs.

I need to do more work on the test dashboard too; since we test the whole history it shouldn't be hard to add a "failing since n commits ago" column.