CodeRabbit Review adds semantic diff view

This is a sensible quality-of-life upgrade, but it also reveals how far traditional diffs are from the way code review actually works in agent-heavy repos.

• –Semantic grouping reduces the false alarms created by refactors, moves, and formatting churn, which is where a lot of review time gets wasted
• –Token-level highlighting is the right compromise: it preserves precision without forcing reviewers to mentally reconstruct the change
• –The feature fits CodeRabbit’s broader pitch as a review layer for bigger PRs, not just a comment bot
• –The timing makes sense for AI-generated code, where mechanical churn is often much larger than the actual logic change
• –This is more infrastructure for understanding code than a flashy new AI model feature, but it is likely to matter more in day-to-day use