`vec_init()` is much slower than `vec_slice()`

[DavisVaughan]

Due to the R level calls to vec_cast() and vec_assert(). We could definitely move these checks down to C

library(vctrs)

x <- rep(list(1), times = 10000)

bench::mark(
  lapply(x, vec_init, n = 1L),
  lapply(x, vec_slice, i = NA_integer_),
  lapply(x, function(xx) {
    .Call(vctrs:::vctrs_init, xx, 1L)
  })
)
#> Warning: Some expressions had a GC in every iteration; so filtering is disabled.
#> # A tibble: 3 x 6
#>   expression                                                         min  median
#>   <bch:expr>                                                    <bch:tm> <bch:t>
#> 1 lapply(x, vec_init, n = 1L)                                    71.01ms 76.23ms
#> 2 lapply(x, vec_slice, i = NA_integer_)                           8.64ms 10.28ms
#> 3 lapply(x, function(xx) { .Call(vctrs:::vctrs_init, xx, 1L) })   8.57ms  9.46ms
#> # … with 3 more variables: itr/sec <dbl>, mem_alloc <bch:byt>, gc/sec <dbl>