When an error happens in `common_setup` it's often difficult to tell what went wrong

Because we run everything in parallel, the error often doesn't propagate all the way up until enough log lines have been emitted for the error to leave the scroll buffer. This can make debugging pretty hard!

Some solutions
- Don't emit to stdout unless an error happens, so we don't flood the scroll buffer
- Emit loglines but also write to a file the user can later inspect
- When an error happens, propagate the error faster and stop execution on other hosts
- If an error happens have the output from the failed command, and print it once the error finally propagates