Adds a comprehensive test suite cli_regression that tests output of certain CLI commands against the old version. The way it works is that it installs a second "golden" bender and diffs the two outputs. The idea of those tests is not to freeze the script output of bender, but rather to give a feedback to the user whether there has been an unwanted regression.

It's also added in the CI as a separate workflow that only triggers with PR. Running it on push to master would not make any sense anyway since it would just compare master against master. Unfortunately, GitHub does not have any allow-failure feature that would just warn users (there is an open feature request since 5 years), meaning this CI workflow might indicate failure which can be ignored (after double-checking the diff output).

Note on windows tests

There is an inconsistency in the script output in windows CI builds. The reason is that during checkout of a repository, git can transform newline characters to platform specific newline characters. For some reason, the git checkout done by the CI and cargo install --git checks out differently, but specifying the newline character in .gitattributes solves this issue.

TODOs:

• Check test list for consistency