Improve behaviour after being down for a while

[mjl-]

When we start up again, we'll forward the log, potentially doing lots of requests tot he sumdb server. We rate limit ourselves (perhaps a bit too quickly), and keeping printing that we are waiting. We can probably relax the rate limiter a bit, and ideally say for how long we will be sleeping, instead of printing a line each second.

We will also read all new modules/versions in memory. If we've been down for a long time, this will cause trouble. We can verify a new tree, but don't have to process all new records in one transaction. We can just handle them in batches, and commit in between. It would still be nice if we only sent emails at the end. So: first insert ModuleUpdates to notify about, in separate batches, and only then send notifications.