Consider what `epi_slide(.window_size = Inf)` should output when min `time_value` differs by epikey

[brookslogan]

suppressPackageStartupMessages({
  library(dplyr)
  library(epiprocess)
})
vctrs::vec_rbind(
  tibble::tibble(geo_value = 1, time_value = 1:4 + 0, value = 1:4),
  tibble::tibble(geo_value = 2, time_value = 3:5 + 0, value = 11:13)
) %>%
as_epi_df() %>%
epi_slide(~ sum(.x$value), .window_size = Inf)
#> An `epi_df` object, 7 x 4 with metadata:
#> * geo_type  = hhs
#> * time_type = integer
#> * as_of     = 2025-04-08 16:57:38.919515
#> 
#> # A tibble: 7 × 4
#>   geo_value time_value value slide_value
#>       <dbl>      <dbl> <int>       <int>
#> 1         1          1     1           1
#> 2         1          2     2           3
#> 3         1          3     3           6
#> 4         1          4     4          10
#> 5         2          3    11          NA
#> 6         2          4    12          NA
#> 7         2          5    13          NA

# (We get the same result with epi_slide_sum; something like this is in our test suite.)
vctrs::vec_rbind(
  tibble::tibble(geo_value = 1, time_value = 1:4 + 0, value = 1:4),
  tibble::tibble(geo_value = 2, time_value = 3:5 + 0, value = 11:13)
) %>%
as_epi_df() %>%
epi_slide_sum(value, .window_size = Inf)
#> An `epi_df` object, 7 x 4 with metadata:
#> * geo_type  = hhs
#> * time_type = integer
#> * as_of     = 2025-04-08 16:57:39.021215
#> 
#> # A tibble: 7 × 4
#>   geo_value time_value value value_running_sum
#>       <dbl>      <dbl> <int>             <dbl>
#> 1         1          1     1                 1
#> 2         1          2     2                 3
#> 3         1          3     3                 6
#> 4         1          4     4                10
#> 5         2          3    11                NA
#> 6         2          4    12                NA
#> 7         2          5    13                NA

^{Created on 2025-04-08 with reprex v2.1.1}

The NAs in the second group are presumably coming from completing time values 1&2 with NAs. Is this what we want? On one hand, it makes the input time_values contributing to each output time_value the same for each geo_value. On the other hand, it makes the result inconsistent with what one might expect from explicitly spelling out edf %>% group_by(geo_value) %>% epi_slide(....) %>% ungroup(), i.e., that it'd be the same as group-splitting/mapping and performing the same operation, and recombining. (We might have some other lesser violations of this expectation with period-inference somewhere, maybe epix_slide, but in general I think we've been following this as well. [Another violation is in handling of explicit .ref_time_values; if we split out into geos with partial ref time availability then we would raise an error.)