Prevent `strwidth()` and `par()` calls from leaking graphics devices

[nanxstats]

Fixes #227

This is new try (after PR #228) to fix the issue reported in #227 where calls to graphics::strwidth() and graphics::par() unintentionally generate Rplots.pdf files in some environments due to opening a new default graphics device that is never closed.

The discussion in that PR was very helpful in getting this "more proper" solution.

• Add a with_graphics_device() helper to run graphics::strwidth()/graphics::par() without implicitly leaving the default device open by opening a null-output device only when needed and closing it on exit.
• Refactor code under R/ and tests/ to use the helper.

Performance comparison:

microbenchmark::microbenchmark(
  r2rtf:::with_graphics_device(graphics::strwidth("Hello world!", units = "inches", cex = 1, font = 1, family = "sans")),
  graphics::strwidth("Hello world!", units = "inches", cex = 1, font = 1, family = "sans"),
  times = 10000
)

#>   min    lq     mean median    uq    max neval cld
#> 4.305 4.715 5.132995  5.002 5.371 33.579 10000   a
#> 1.968 2.214 2.418569  2.337 2.501 26.855 10000   b

Note that changing this might affect some strwidth() calculation results slightly.