group_by() sorts NA first

``` r
library(dplyr, warn.conflicts = FALSE)

tbl <- tibble(
  x = c("apple", NA, "banana"), 
  y = 1:3
)
tbl %>% 
  group_by(x) %>% 
  group_data()
#> # A tibble: 3 x 2
#>   x      .rows    
#>   <chr>  <list>   
#> 1 <NA>   <int [1]>
#> 2 apple  <int [1]>
#> 3 banana <int [1]>
```

simplified from #4220