According to https://github.com/actions/runner/issues/987, this should run the post step (cleanup.js) also when a workflow fails.
Probably most important on self-hosted runners that are not ephemeral, to terminate SSH agents from failed jobs as well.
Thanks to @thommyhh for this contribution!
Unless the `SSH_AUTH_SOCK` is configured explicitly, this change will make the SSH agent use a random file name for the socket. That way, multiple, concurrent SSH agents can be used on non-ephemeral, self-hosted runners.
A new post-action step will automatically clean up the running agent at the end of a job.
Be aware of the possible security implications: Two jobs running on the same runner might be able to access each other's socket and thus access repositories and/or hosts.
