perf(root-search): optimize command scoring

[JustYannicc]

What changed

• Added scripts/test-root-search-perf.mjs, a focused synthetic root-search perf harness that compares the legacy two-pass command candidate output against the optimized path.
• Reused the root command match kind/score from rankCommands instead of rescanning the same command fields in useLauncherCommandModel.
• Skipped the legacy filterCommands list for normal non-empty root searches, where display already comes from assembled ranked root sections. Calculator fallback queries still compute the legacy list.

Why

• Root search was doing duplicated O(commands * fields * query terms) work on each non-empty query.
• The legacy filtered list is not the displayed source for root search results, and command candidates only need the final root scoring pass that feeds scoreRootSearchCandidate.

Compatibility impact

• Intended behavior is unchanged: the perf harness asserts optimized command matches have the same command ids, match kinds, and match scores as the legacy two-pass candidate path for representative queries.
• No user-visible strings or i18n keys changed.
• Returned legacy list fields are still populated for empty queries and calculator fallback queries; normal root-search display continues to use ranked root sections.

How tested

• node scripts/test-root-search-ranking.mjs
• node scripts/test-root-search-perf.mjs
• node --test 'scripts/test-*.mjs'
• node scripts/check-i18n.mjs
• ./node_modules/.bin/tsc -p tsconfig.renderer.json --noEmit (fails on existing repo type errors outside this change, including App.tsx, CameraExtension.tsx, LiquidGlassSurface.tsx, browser profile typing in useLauncherCommandModel.ts, raycast runtime files, and quicklink icon typing).

Perf numbers from node scripts/test-root-search-perf.mjs over 6,000 synthetic commands, 8 queries, 6 iterations:

• Pre-edit captured baseline: legacy filter + two-pass command scoring median 3570.81ms; optimized estimate before score reuse 2190.89ms.
• Post-edit focused run: legacy comparator median 3872.38ms; optimized command scoring path median 1753.83ms; speedup 2.21x.
• Post-edit full node --test run: legacy comparator median 3747.75ms; optimized command scoring path median 1661.77ms; speedup 2.26x.

[JustYannicc]

Superseded by the consolidated upstream PR #535: #535