Test formatter changes against baseline format

[straight-shoota]

For testing formatter changes we previously skipped code bases that were not already properly formatted because they would report formatter errors unrelated to the particular changes being tested.

But we actually don't need the codebase to be pre-formatted. We can run a baseline formatter first and then see if there are any changes against that.

This patch introduces a second Crystal compiler that serves as the baseline (defined by $CRYSTAL_BASE).
If a baseline compiler is available, we first run $CRYSTAL_BASE tool format and then only consider changes that $CRYSTAL tool format produces against that baseline.

CI installs the baseline compiler in /tmp/crystal-baseline/.

This allows us to run formatter tests against any code base, regardless of whether it's already well-formatted or not. This increases the potential surface area a lot.
At this point we can probably extract the individual formatter test definitions from the bats files and just iterate a long list of repository URLs.

[ysbaddaden]

It seems to be broken?

[straight-shoota]

No idea what's going on with athena. Looks like a bug in shards? 🤔

The failure in admiral.cr seems to be a formatter bug in the baseline which produces invalid code: crystal-lang/crystal#16755

[ysbaddaden]

Yeah, Shards is trying to fetch athena-framework/spec.git twice:

I: Fetching https://github.com/athena-framework/spec.git
...
I: Fetching https://github.com/athena-framework/spec.git
E: Failed to clone https://github.com/athena-framework/spec.git: Failed git clone -c core.askPass=true -c init.templateDir= --mirror --quiet -- https://github.com/athena-framework/spec.git /home/runner/.cache/shards/github.com/athena-framework/spec.git.
fatal: destination path '/home/runner/.cache/shards/github.com/athena-framework/spec.git' already exists and is not an empty directory.