Generalize review personas and add agent-process eval canaries

[JamesCarnley]

Summary

The current review personas are useful, but they still duplicate shared review protocol and are somewhat overfit to the PIN/TAG migration.

This issue tracks the next cleanup pass so the review system ages well and can be tested more reliably.

Scope

• De-duplicate shared persona boilerplate
• Generalize the default reviewers beyond the PIN/TAG migration
• Add an eval / canary suite for the agent process

Desired work

Persona cleanup

• move shared review-format / verification-context / preflight rules into a common base pattern
• keep persona files focused on role-specific heuristics
• reduce repetition across review-squad, persona README, and persona prompt files
• replace hardcoded migration-specific reads in the general reviewers with a pluggable "governing docs for this change" slot

Agent-process eval canaries

Create a small canary suite for prompts like:

• add a field to the DATA schema
• backfill append-only index entries
• rename a debug UI label
• change a TS API the Vite client consumes
• post a PR review comment
• approve your own PR

The goal is to verify that the documented workflow actually fires the intended behaviors on new models and prompt revisions.

Why

This was identified after PR #10. The process works, but it still depends too much on prompt discipline and duplicated instructions.

References

• docs/review/personas/README.md
• docs/review/review-squad.md
• docs/agent-workflow.md
• docs/FUTURE_WORK.md → Tooling & Process