Data objects need (optional) confidence intervals

[eschrom]

Cumulative and incident uptake data objects (whether they represent input data or model projections) should have not only point estimates, but also columns for upper and lower bounds of credible intervals.

[swo]

• This will become relevant when we do Bayesian models that need input data error distributions
• We probably want some validation but not super-hard because the data API should take care of that mostly