support unnest_wider on Date class columns

[atusy]

unnest_wider on Date class columns fail unless simplify = FALSE is set.

df = tibble::tibble(x = list(c(a = as.Date("2020-01-01"), b = as.Date("2020-01-02"))))
tidyr::unnest_wider(df, x)
#> Error in as.Date.default(value): do not know how to convert 'value' to class "Date"
tidyr::unnest_wider(df, x, simplify = FALSE)
#> # A tibble: 1 x 2
#>   a          b         
#>   <date>     <date>    
#> 1 2020-01-01 2020-01-02

^{Created on 2020-08-05 by the reprex package (v0.3.0)}

As far as I investigated, the issue comes from this line:

tidyr/R/rectangle.R

Line 404 in 61e9209

x[n == 0] <- list(NA)

Following reprex shows that list(NA) cannot be converted to class "Date" whereas NA can be.

x = c(a = as.Date("2020-01-01"), b = as.Date("2020-01-02"))
x[1] <- list(NA)
#> Error in as.Date.default(value): do not know how to convert 'value' to class "Date"
x[1] <- NA

^{Created on 2020-08-05 by the reprex package (v0.3.0)}

In this PR, I suggest substitute elements by list(NA) if x is a list, and by NA otherwise.

[atusy]

It seems like rectangling Date class had been supported before 4b4e516

[hadley]

Possibly related to #1089

[hadley]

I took a slightly different approach — I don't understand exactly why, but in this case x isn't a list, so we don't need to attempt simplification at all.