Fix detection of change in dependency

[DavisVaughan]

This should fix the failing main branch

In #1789, support was added for the imports field of use_standalone() and internally that used use_package() to add the required packages to Imports. use_package() called use_dependency(), and I think use_dependency() had a small bug in it, which I'll explain below.

The changes caused this test to fail because of snapshot changes

usethis/tests/testthat/test-tidyverse.R

Lines 15 to 23 in a336c67

	test_that("use_tidy_dependencies() isn't overly informative", {
	skip_if_offline("github.com")
	create_local_package(fs::path_temp("tidydeps"))
	use_package_doc()
	withr::local_options(usethis.quiet = FALSE)
	expect_snapshot(use_tidy_dependencies())
	})

Specifically it was adding this to the snapshots

Execution halted
  "  v Adding '@importFrom lifecycle deprecated' to 'R/tidydeps-package.R'"
  "  v Writing 'NAMESPACE'"
  "  v Writing 'R/import-standalone-purrr.R'"
+ "  * Refer to functions with `rlang::fun()`"

But that is weird, because use_tidy_dependencies() already called use_package("rlang") up above, and its usage of use_standalone("r-lib/rlang", "purrr") should not have caused a "change" in package dependency type that should trigger this how_to_use() line in use_package(), but it obviously does:

usethis/R/package.R

Lines 40 to 43 in a336c67

	changed <- use_dependency(package, type, min_version = min_version)
	if (changed) {
	how_to_use(package, type)
	}

For reference the rlang standalone file does not use a versioned version of rlang
https://github.com/r-lib/rlang/blob/9b450274026baa46aa8f4520fdefd87de02e4565/R/standalone-purrr.R#L6

Turns out use_dependency() was actually returning TRUE here by mistake, because there isn't a delta == 0L branch that exits and returns FALSE (i.e. the case of package exists and no change in version). I've tweaked use_dependency() to have a changed variable that keeps track of whether or not something has changed in the DESCRIPTION file (i.e. changed <- TRUE always follows a desc$write() call).

This change results in a snapshot change that I actually think is a bug fix, because now 2 calls to use_package("withr") results in the 2nd call being completely silent

---

I also sort of question the usage of use_package() in use_standalone() over the quieter use_dependency(), but that is a separate issue, since this one is clearly a bug.

[jennybc]

I had already sorted the messaging thing out (b9c4b88), but yes this looks like an improvement in the return value.