population scaling with no matching names silently returns only `NA`s

using e.g. this step
```
preproc %<>% step_population_scaling(
all_numeric(),
df = state_census,
df_pop_col = "pop",
create_new = FALSE,
rate_rescaling = 1e5,
by = c("geo_value" = "abbr")
)
```
with this "data"
```
constant <- as_epi_archive(tibble(
geo_value = "g1",
time_value = seq(as.Date("2012-01-01"), by = "day", length.out = 90),
version = seq(as.Date("2012-01-01"), by = "day", length.out = 90),
a = 25
))
```
is happily passed on to `lm.fit`, which then sees no data to fit on. Would probably be good to error out if there are columns in the dataset that aren't in `df_pop_col`.