Standardize how LLM output is stored in a ProviderResponse output object

[efung]

ProviderResponse has an output field that is documented to be either a string or an object.

If it's a string, then the Web UI shows a nice diff between the LLM outputs when comparing two prompts or providers.

However, if it's an object, the serialized object is shown and you lose the nice diff output.

Can a convention be established about what goes into the object, specifically if a field with a specific key exists, then that represents the LLM output? (And any other fields can be visible in the Web UI if you click on the magnifying glass).

[typpo]

This seems reasonable. What do you think we should call the displayed field? content, display, text?

[efung]

This seems reasonable. What do you think we should call the displayed field? content, display, text?

I prefer text or content.

[typpo]

Added a frontend configuration for this in #893