Improve `epi_slide` error messaging on bad `ref_time_values`

[brookslogan]

When the user inputs ref_time_values before or after the time_values in the epi_df (even if there'd be data in the window), it gives a hard-to-understand error message:

Error: Can't subset columns that don't exist.
✖ Column `slide_value` doesn't exist.
Run `rlang::last_error()` to see where the error occurred.

We should instead detect this case and output an understandable error message.

Examples:

edf = bind_rows(
  tibble(geo_value = "ak", time_value = as.Date("2020-01-01") + 1:200, value=1:200),
  tibble(geo_value = "al", time_value=as.Date("2020-01-01") + 1:5, value=-(1:5))
) %>%
  as_epi_df()

f = function(x, ...) tibble(value=mean(x$value), count=length(x$value))

## --- These cases generate the error: ---

## `ref_time_values` + `align` that result in no slide data, generate the error:
edf %>% group_by(geo_value) %>% epi_slide(f, n=3L, ref_time_values=as.Date("2020-01-01")) # before the first, no data in the slide windows
edf %>% group_by(geo_value) %>% epi_slide(f, n=3L, ref_time_values=as.Date("2020-01-01")+207L) # beyond the last, no data in window

## `ref_time_values` + `align` that should? have slide data, but generate the error:
edf %>% group_by(geo_value) %>% epi_slide(f, n=3L, ref_time_values=as.Date("2020-01-01"), align="left") # before the first, but we'd expect there to be data in the window
edf %>% group_by(geo_value) %>% epi_slide(f, n=3L, ref_time_values=as.Date("2020-01-01")+201L) # beyond the last, but still with data in window

## --- This case doesn't generate the error: ---

## (this doesn't produce an error; the error appears only if the ref time values are out of the range for every group:)
edf %>% group_by(geo_value) %>% epi_slide(f, n=3L, ref_time_values=as.Date("2020-01-01")+200L)