fake stream output if model doesn't support it but stream=True

[krrishdholakia]

[krrishdholakia]

if somebody used us assuming we'd return a stream (generator object i believe) and we returned a dict, we'd break in code for them.

[zakhar-kogan]

Does this need a separate model flag, e.g., supports_streaming = True, to distinguish?

[krrishdholakia]

Not really - we already have a customstreamwrapper, we can just add this as default logic and send all model results there if stream=true

[krrishdholakia]

done