Delay dplyr_quosures() auto naming until needed

[DavisVaughan]

This mostly affects filter(), because expressions are never named there but we have been auto naming them anyways.

library(dplyr)

df <- tibble(x = 1, y = 2:3)

bench::mark(filter(df, x == 1, y == 2), iterations = 10000)

# Before
#> # A tibble: 1 × 6
#>   expression                      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>                 <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 filter(df, x == 1, y == 2)    987µs   1.14ms      847.    1.35MB     13.7

# After
#> # A tibble: 1 × 6
#>   expression                      min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>                 <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 filter(df, x == 1, y == 2)    653µs    835µs     1218.    1.75MB     12.9

There are some benefits for mutate() and summarise(). If an across() expansion occurs, then we previously had auto-named the expansion for nothing. This is less important though.

[DavisVaughan]

The fact that we capture and store the original quosure is somewhat important here, as there are cases with pick() where we expand the quosure into something else, so we need a reference to the original one to be able to auto-name it correctly. I've added a test for this.