Another fix re: version numbers

[jennybc]

Related to #1869

From an email from Kurt Hornik to many package developers:

A recent PR (https://bugs.r-project.org/show_bug.cgi?id=18548) showed
that whereas version inputs to numeric_version() and package_version()
were always documented to be "suitable character vectors" (or numeric
vesion objects), the implementation also accepted numeric inputs via
as.character() coercion, the results of which however are (at least
currently) dependent on the setting of the OutDec option. Given that
numeric input is certainly error-prone, imagine

as.numeric_version(1.20)
as.numeric_version(1.5-1)

R Core has decided it would be best to gradually phase out such inputs.

Thus, With c84610 R-devel now warns about such inputs, which will
typically not show in the check results (but immediately giving errors
would cause too much disruption). Some of these warnings can be found
by code analyis. I list these uses below, can you please fix as
necessary?

When doing so, please check the check outputs for such warnings, e.g. by

grep -r 'invalid non-character version specification' PKGNAME.Rcheck

(One can for now set env var R_CALLS_INVALID_NUMERIC_VERSION to
something non-empty to get the call stacks for the warnings; this may
need increasing option warning.length from its default.)

Please fix before 2023-07-12.

The section given for usethis:

$usethis
$usethis$`usethis/tests/testthat/test-author.R`
$usethis$`usethis/tests/testthat/test-author.R`[[1]]
getRversion() < 4


$usethis$`usethis/tests/testthat/test-tidyverse.R`
$usethis$`usethis/tests/testthat/test-tidyverse.R`[[1]]
getRversion() < 3.2

$usethis$`usethis/tests/testthat/test-tidyverse.R`[[2]]
getRversion() < 3.2