background_updates was using `call_later` in a way that leaked the logcontext
into the reactor.
We could have rewritten it to do it properly, but given that we weren't using
the fancier facilities provided by `call_later`, we might as well just use
`async.sleep`, which does the logcontext stuff properly.
A bit of a cleanup for background_updates, and make sure that the real
background updates have run before we start the unit tests, so that they don't
interfere with the tests.
user_ips is kinda big, so really we want to add the index in the background
once we're running. Replace the schema delta with one which will do that.
I've done this in a way that's reasonably easy to reuse as there a few other
indexes I need, and I don't suppose they will be the last.
The progress for each background update is stored as a JSON blob in the
database. Each background update is broken up into separate batches.
The batch size is automatically tuned to try avoid blocking single
threaded databases for too long.
