Cheap/Expensive Stage Routing for a Review Pipeline

Design a multi-stage code review pipeline for pull requests in a large repository, where the goal is to catch real issues (not just style nits) while keeping cost proportional to PR size rather than running every PR through the most expensive model regardless of how small or trivial it is.

1. Describe the stages of your pipeline and what each stage is responsible for.
2. For each stage, specify which model it should run on and justify the choice using the cheap-stage/expensive-stage principle from this subject.
3. Explain concretely, using the subagent model-resolution precedence chain, how you would implement your model choices so that an individual reviewer can't accidentally run the whole pipeline on an expensive model by mistake, while still allowing a deliberate one-off exception for a specific high-stakes PR.