Shell-command should reuse previously computed outputs of tasks that errored in output processing

[tclose]

What would you like changed/added and why?

Would be nice if errors in output interfaces of shell-command did not require you to rerun the command again to debug. We should be able to detect that the error has just occurred in the output, not invalidate the cache, and instead just pick up again at the start of the output processing.

What would be the benefit?

Would make debugging shell command output errors much quicker and less painful, especially for long-running commands.