Idea: Allow garbo to return multiple responses for each prompt if there are unclear results

[Greenheart]

Perhaps we could allow garbo to return multiple options for the same task?

If we transform each schema into a Zod array (z.array()), update the prompts, and then allow garbo to return one or more results with the same structure. Then a human could pick the best option via a validation UI.

This might help to solve some issues.