Refactor titleGrob()

[teunbrand]

This PR aims to fix a performance bottleneck for #5030 as identified in #5030 (comment).

Briefly, it introduces the following changes:

1. The titleGrob grobs no longer have viewports.
2. Instead of viewports, the position of the textGrob is adjusted to get the same effect.
3. The titleGrob() constructor now does the job of what previously was done by title_spec() and add_margins().
4. The assemble_strip() no longer needs to awkwardly hassle with all these viewports.
5. A bunch of functions, like title_spec(), add_margins(), margin_width(), margin_height() have become obsolete. I've removed them, but we might expect this to bite us during revdepchecks because they're missing.

Benchmark below shows a ~4 fold reduction in drawing time for 574 titleGrobs.

library(ggplot2)
library(grid)
library(scales)

x <- rescale(as.numeric(economics$date))
y <- rescale(as.numeric(economics$unemploy))
lab <- as.character(economics$pop)

make_grob <- function(fun, x, y, label, margin = margin()) {
  do.call(
    grobTree,
    Map(
      fun,
      x = x, y = y, label = label, hjust = 0.5, vjust = 0.5,
      MoreArgs = list(margin_x = TRUE, margin_y = TRUE, margin = margin)
    )
  )
}

grob_old <- make_grob(
  ggplot2:::titleGrob, x = x, y = y, label = lab, 
  margin = margin(5, 5, 5, 5)
)

devtools::load_all("~/packages/ggplot2/")
#> ℹ Loading ggplot2

grob_new <- make_grob(
  titleGrob, x = x, y = y, label = lab, 
  margin = margin(5, 5, 5, 5)
)

render <- function(grob) {
  file <- tempfile(fileext = ".png")
  ragg::agg_png(file)
  grid.newpage(); grid.draw(grob)
  dev.off()
  unlink(file)
}

bench::mark(
  render(grob_old),
  render(grob_new),
  min_iterations = 10
)
#> Warning: Some expressions had a GC in every iteration; so filtering is disabled.
#> # A tibble: 2 × 6
#>   expression            min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>       <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 render(grob_old)    575ms    585ms      1.67    3.67MB     5.02
#> 2 render(grob_new)    126ms    135ms      7.50  132.94KB     5.25

^{Created on 2023-04-17 with reprex v2.0.2}

[thomasp85]

LGTM... Can I get you to run the visual tests on a non-arm64 computer (assuming you are working on a M1 Mac). I've found the chip architecture influences these weird rounding issues and all the large theme tests never matches on my machine

[teunbrand]

Can I get you to run the visual tests on a non-arm64 computer

These visual tests were run on a windows machine with an x64-based processor. Bit weird that CPU architecture matters for this rounding.