revert back to retaining NULL instead of coalescing empty list-col elements into nested NAs

[egouldo]

Replacing NULLs with NAs results in the missing vals being a PITA to work with... can't call fns like drop_na() or easily filter out missing elements of the list-column.

Also, from replace_na():

# Replace NULLs in a list: NULLs are the list-col equivalent of NAs
df_list <- tibble(z = list(1:5, NULL, 10:20))
df_list %>% replace_na(list(z = list(5)))
#> # A tibble: 3 × 1
#>   z         
#>   <list>    
#> 1 <int [5]> 
#> 2 <dbl [1]> 
#> 3 <int [11]>

so I think we should stick with what the equivalent is for list-columns as above.