Diffs adds CodeView for giant diffs

This is a good example of a “boring” infra problem turning into product differentiation: if your diff viewer fails at scale, the whole review workflow feels broken.

• –`CodeView` combines virtualization, inverse sticky behavior, and scroll anchoring to keep giant diffs readable even when JavaScript falls behind
• –The layout system uses estimates first, then corrects with measured deltas, which is the right tradeoff for keeping large reviews fast
• –The memory work matters: parsing huge patches can retain far more memory than expected, so detaching strings is a practical fix, not an academic one
• –This is especially relevant for AI-generated PRs, where reviewers are increasingly dealing with larger, multi-file change sets
• –The DiffsHub demo is smart packaging: it turns a technical component into something people can immediately stress-test in the browser