count()/add_count() should preserve input classes

[yutannihilation]

(Originally reported at #4051 (comment).)

``` r library(dplyr, warn.conflicts = FALSE)

x <- as_tibble(iris)
g <- group_by(x, Species)
class(g) <- c("myclass", class(g))

x %>%
add_tally() %>%
class()
#> [1] "tbl_df" "tbl" "data.frame"

x %>%
add_count() %>%
class()
#> [1] "tbl_df" "tbl" "data.frame"

<sup>Created on 2019-01-07 by the [reprex package](https://reprex.tidyverse.org) (v0.2.1)</sup>
</details>

[hadley]

Minimal reprex:

library(dplyr, warn.conflicts = FALSE)

x <- data.frame(x = 1:3)
x %>% count() %>% class()
#> [1] "tbl_df"     "tbl"        "data.frame"
x %>% add_count() %>% class()
#> [1] "tbl_df"     "tbl"        "data.frame"

<sup>Created on 2019-12-31 by the reprex package (v0.3.0)</sup>

(tally() and add_tally() don't touch the classes so don't need to be considered further)

[hadley]

We might want to solve this with a generic mechanism for temporarily grouping, performing some operation, and then restoring groups (some sort of with_groups()), which we could also export.

[hadley]

Somewhat more comprehensive reprex:

df <- structure(data.frame(x = 1:3), class = c("myclass", "data.frame"))
df %>% count() %>% class()
df %>% count(x) %>% class()
df %>% add_count() %>% class()
df %>% add_count(x) %>% class()