Handle potential 0-segment regimes

[noah10]

In #352 we addressed the question of what to do when a regime has only one segment and therefore a confidence interval can't be calculated. While it's very unlikely, we should also handle the case for when a regime has 0 segments. Right now, that would be kicked out because we calculate performance statistics only for practiced regimes, that is, regimes that have associated segments. We should not kick them out.