Updater integration tests

[fenhl]

Updater bugs like #44 are especially bad because fixing them doesn't help users who get stuck on a broken version. The release script should include a test which tries to update from the version being released to itself, to make sure nothing is broken. This test would necessarily have to run after the release itself is published, so it wouldn't prevent these bugs from being released, but it would avoid the scenario where a bug isn't noticed until the next version is released.