feat: Extends performance domain with measurement skills

[MajorLift]

Motivation

domains/performance already has five skills, and none of them measures anything.

The performance skill is mobile-only. The four perf-* skills are extension-only pattern guides, and testing/performance-testing runs mobile E2E.

So the repo tells you what to change, per platform, and nothing tells you whether the change worked. Three gaps follow, and each has a specific cost:

The extension has no profiling skill. Every profiling reference on main is mobile tooling that does not exist in a browser. An extension engineer asking "where is the time going" has nothing to open.

Nothing proves a render optimization landed. A useMemo that fires on every render still looks correct in review and still ships. The mechanism is plausible either way. Only the re-render count separates a working useMemo from one that fires every render, and a delivery gate is what stops a change shipping on the strength of the story.

Nothing says what a benchmark difference means. performance-testing runs the benchmark. How many rounds, how much variance, and whether a 4% gap is a result are all unanswered. Without that a benchmark reports noise as a win, which is worse than not running one, because it produces a number people cite.

Web Vitals gets its own knowledge for the same reason: production and benchmark numbers routinely disagree, and the disagreement is a property of how they are collected rather than a regression.

Overview

Four skills for measuring frontend performance. The antipattern scans are in #43.

• extension-profiling compares two branches with why-did-you-render, the React DevTools Profiler and E2E benchmarks.
• react-render-delta proves a memoization change reduced work, using re-render counts, reselect's .recomputations(), A/B arms toggled at a flag, and a delivery gate with a reported band.
• benchmark-design covers session hygiene, per-round reporting and artifact grouping.
• data-analysis runs collection → filtering → curation → questioning → synthesis, with confidence assigned per finding.

Knowledge: metrics-pipeline-design, benchmark-statistical-hygiene, and three on Web Vitals: runtime metrics, attribution import, and why production numbers disagree with benchmark numbers.

Showcase

No trial run yet. These are measurement instruments, and the honest evidence for one is a run that moved a decision. There is not one to cite.