minPostCombinationDuration filters all eras, not just combination spillovers

[katy-sadowski]

Describe the bug
I think it's here, per Copilot :)

TreatmentPatterns/R/constructPathways.R

Line 802 in fb4437f

dplyr::filter(.data$durationEra >= minPostCombinationDuration | is.na(.data$durationEra))

To Reproduce
Run on data containing eras that are less than minPostCombinationDuration but don't overlap any other / aren't contributing to a combination. They're being dropped.

Expected behavior
minPostCombinationDuration operates per documentation: "Minimum time an event era before or after a generated combination treatment should last to be included in analysis"

I know there is a Best Practice recommendation to make minPostCombinationDuration less than minEraDuration, to prevent this silent failure mode, the recommendation should either be enforced via a constraint in the function, or the bug should be fixed.

There is a use case to have minEraDuration and minPostCombinationDuration be different. We may not expect all rx to stop at exactly the same time in a combo while also wanting to count all rx starts regardless of length.