Allow "nonexistent version" test to work without brew

[piperswe]

Currently, this test fails if Homebrew isn't present as it expects a
Homebrew-specific output. This commit varies the expected output based
on whether Homebrew is installed.

This patch was originally created for the Debian package for ruby-build:
https://salsa.debian.org/ruby-team/ruby-build/-/blob/607093e5198cc75ffa8e9a3b551843be5f2f6d93/debian/patches/0001-Fix-nonexistent-version-test-to-work-regardless-of-e.patch

[mislav]

Hi, thank you for offering to make the test more stable. I'm unclear about the original failure on systems without brew, though. The test scenario in question has this line:

stub_repeated brew false

This creates a test-specific stub for the brew binary for the duration of the test. This stub is dynamically injected into PATH in test mode. So, for the purpose of the test it should not matter whether brew exists on the user's system or not. Our brew stub should take precedence.

So if this test fails on systems without brew for anyone, we shouldn't aim to fix a single test scenario and its output, but to fix the stub_repeated brew system that's used in tests. Thoughts?

This patch was originally created for the Debian package for ruby-build:

When was this patch created? Did you author it? Have you experienced the test failure in question?

[piperswe]

Turns out I had forgotten the context - the patch is a few years old at this point. It's not actually anything to do with Homebrew, and I can remove that part. The Debian build environment doesn't have the .gitdirectory. I'll update this PR to only check for the existence of .git - I think it's reasonable to want this test to pass when you download the tarball from GitHub instead of cloning.