[Feature] Run two models, and have the AI compare the results, and report any differences

[michaelbeijer]

Summary

Suggested automated workflow:

• Run Sonnet as the primary translation
• Run GPT in parallel
• Use QA report approach – automated comparison catches the divergences, and divergences are exactly where the interesting decisions lie
• A human reviewer then adjudicates the divergences