More comprehensive integration/performance tests

We should be able to test slotmachine with some representative (but probably completely anonymised) real-world data, so we can easily do performance benchmarking. Ideally this would be both a fresh schedule run and an update.

We probably also want a mid-sized test dataset which is larger than the current tests, but quicker to run than a full event.