Consider supporting `epi_slide_opt(fill =)`

[brookslogan]

We don't properly/meaningfully use data.table::froll* fill = args in epi_slide_opt. #611 raises an error if a user tries. One might want to use this, e.g., to figure out which rows didn't have non-NA-with-no-fill = results due to having padding as part of their computation window, vs. due to a time gap or explicit NA.

We can't just pass fill = along for two reasons:

• the values set to fill are all associated with non-.real time_values introduced by edge-padding
• after >= 1 case needs special treatment

Marking this low priority initially until we hear that it'd be useful.