selectFromDB needs to report out if whole encountered_weeks are being dropped with NaNs

[famulare]

We just caught an issue that is important for model reliability. selectFromDB(..., na.rm=TRUE) silently drops rows with NaN covariates because INLA will just drop them anyway. This makes sense if NaNs are due to randomly missing data. However, when whole blocks of queried data have NaNs, we need at least a warning, if not an outright error. When that happens, something is either wrong with the request or the data.