`clear_lastminute_nas` will erase too much data as written for `chng`

[dsweber2]

Currently, clear_lastminute_nas will remove any row which has any NA's in it whatsoever. With multiple signals, this can mean removing rows with real data (as an example, see 11-29-24, for which 10-02-24 is the last day with data). Before using chng, we will need another scheme to handle differently missing data. Ideas:

• locf NA's in non-outcome signals. This is not a great idea, as lags likely become meaningless.
• do the equivalent of extend_ahead but for lags (e.g. if the last observation for chng is 30 days behind, adjust the lags of c(0,7,14) to c(30,37,44).
• ?

@brookslogan may have other thoughts I forgot about.