Measure effect of day vs. file aggregation strategy

[matt-har-vey]

Specifically, the rates lambda and omega are expected a priori to vary with time of day, so defining "surveys" (# = nsurveys) as 15-minute files could make the posteriors of these rates multimodal.

It's not clear how / whether this would affect the estimates.

We could add an aggregation period argument (file / day) to the ML data tabulation API and do a controlled experiment with both of its settings.

[matt-har-vey]

Blocked by #78