fix(perf): eliminate the main-thread App Hangs (Sentry triage)

[caezium]

What

Every unresolved Burrow Sentry issue is an AppHang (main thread blocked ≥2 s) — zero real crashes. The 40 issues are ~5 root causes wearing different top-frame hats; the two dominant buckets (BURROW-1 layout, BURROW-N generic app.run()) both regressed after a prior 0.7.1 resolve. This PR fixes the structural root causes.

Changes

A. App icons off the main thread — StatusView.swift (BURROW-R, BURROW-T)

AppIcon.image(for:) walked NSWorkspace.runningApplications on the main thread on every cache miss, once per process row, each 2 s refresh — hundreds of O(running-apps) walks. Now:

• AppIcon.resolve(for:) runs inside the existing off-main process pass, walking the app list at most once and indexing it (O(apps + procs), not O(apps × procs)).
• Views read cachedImage(for:) — a pure, lock-guarded cache read; glyph fallback until the icon lands.
• The popup's handful of rows use image(for:) = cache read + async off-main fill.
• The main thread never walks the app list again.

B. Bounded process table — StatusView.swift, MoleStatus.swift (BURROW-1 + layout tail)

The table's ForEach iterated the full process set (hundreds), rebuilding/diffing every identity each tick and forcing repeated sizeThatFits over the nested ScrollViews (the ScrollViewLayoutComputer → placeChildren recursion in the trace). Now:

• ForEach is capped to the top 100 rows by the current sort, with a "Show all (N more)" affordance.
• ProcRow has a fixed .frame(height: 30) so LazyVStack skips per-child measurement and the ScrollView size cache stays stable.
• ProcessInfo is now Equatable so unchanged rows don't re-render.

C. Throttled clean/optimize report — OperationFlow.swift (BURROW-1G, BURROW-1F)

OperationFlow (@MainActor) re-parsed the whole accumulated transcript (parseTaskReport/mergeSummaryFields) on the main actor for every streamed line — O(n²) over a long run. The live re-parse is now throttled to ~4×/s; terminal events still do a final, authoritative reduce, so the result screen is never stale.

Audited — no change needed

• Modals (BURROW-1K): every NSAlert.runModal() already routes through runModalQuiet(), which pauses Sentry ANR tracking for user-paced modals.
• Blocking waits: the disk-scan (group.wait) and CLI-process (waitUntilExit) calls run off-main.
• Sparklines (BURROW-1M): MiniChart already caps to 120 drawn points; this fingerprint is downstream of the main-thread pressure that A/B/C relieve.
• The framework-frame singletons (CharacterSet, swift_slowAlloc, _platform_memmove, chkstk, AGGraph…) are symptoms of a busy main thread and should shrink as a side effect — not worth chasing individually.

Test plan

• xcodebuild Debug build succeeds, no new warnings.
• Hand-test: open the window with many processes; confirm scrolling/refresh stays smooth, "Show all" works, icons appear.
• Instruments (Hangs + Time Profiler): confirm main-thread time per 2 s tick drops sharply.
• Run a long mo clean and confirm the live report still updates and the final result is correct.

Notes

• ⚠️ Do not auto-resolve the Sentry issues until a release soaks — the top two regressed after a prior resolve; let production confirm.
• Branched off origin/main. Companion PR (customizable menu-bar metrics, 关于菜单栏可以自定义指标的建议 #82) to follow.