Use new_data_frame() in new_tibble()

[DavisVaughan]

Closes r-lib/vctrs#892

I've reworked new_tibble() on top of new_data_frame(). It is faster now than before in some cases, partly due to avoiding setdiff() for class when we don't need it.

library(tibble)

# empty
y <- list()
y_nrow <- 0L

bench::mark(
  tbl = new_tibble(y, nrow = y_nrow),
  iterations = 100000
)

# master
#> # A tibble: 1 x 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 tbl          11.7µs   16.8µs    56017.    27.9KB     61.7

# this PR
#> # A tibble: 1 x 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 tbl          8.76µs   11.5µs    77859.    32.9KB     17.9

x <- unclass(mtcars)
nrow <- 32L

bench::mark(
  tbl = new_tibble(x, nrow = nrow),
  iterations = 100000
)

# master
#> # A tibble: 1 x 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 tbl          10.8µs   14.2µs    69187.        0B     27.0

# this PR
#> # A tibble: 1 x 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 tbl           9.4µs   12.6µs    78395.        0B     15.7

bench::mark(
  tbl = new_tibble(x, nrow = nrow, foo = "bar", foo2 = 1, foo3 = "x"),
  iterations = 100000
)

# master
#> # A tibble: 1 x 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 tbl          12.5µs   16.9µs    59008.        0B     27.7

# this PR
#> # A tibble: 1 x 6
#>   expression      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 tbl          12.7µs   17.2µs    57369.        0B     18.4

[krlmlr]

Thanks. I've arrived to a similar code when trying to use new_data_frame(). At this time I feel it's too much extra code, I think we can do much better with an improved version of new_data_frame() (r-lib/vctrs#892).

If the computation of the "class" attribute becomes an issue, we can move to C.

Happy to take the tests. Do they pass with dev tibble?

[krlmlr]

Thanks!