In c2f4dfc, we're computing the baselines for interaction terms on-the-fly. This is not ideal as it can be done when the result set is created.

•  add a secondBaselineGroup field to DifferentialExpressionAnalysisResultSet
• populate it when we perform the linear model
• backfill values using baselines of individual factors

I'm not sure we allow interaction of continuous factors, but in that case those inherently lack a baseline.

There might be a case when a continuous factor interacts with a categorical one, in which case we should also fill one of the baselines. @ppavlidis is this possible?