feat(desktop): improve hot CPU diagnostics workflow

[huntharo]

Summary

Hot CPU diagnostics are now easier to hand off and safer to operate after a capture. The copied profile text keeps the exact artifact paths, calls out sidecar files, and frames the capture as troubleshooting evidence rather than proof that a bug is still active.

Developer troubleshooting actions now go through diagnostics:* command-bus verbs. The banner can reveal the captured session, Settings can reveal the diagnostics root, and cleanup only targets hot CPU session-shaped direct children under the current hot CPU diagnostics root.

The cleanup path is hardened after review: it uses the same env-aware output root as capture writes, skips active in-process sessions at the main-process command boundary, disables conflicting Settings actions while cleanup is pending, and ignores stale banner reveal failures after a newer profile arrives. Hot CPU sessions also have bounded retention plus a durable runbook for opening .cpuprofile and .heapsnapshot artifacts in Chrome DevTools.

Screenshots

Not included. This is a developer-only diagnostics surface, and the important runtime evidence is the built-app profiling artifact smoke below.

Screen recording

Not included.

How was this tested?

• pnpm exec vitest run --config vitest.workspace.ts apps/desktop/src/main/handlers/__tests__/diagnostics-handlers.test.ts apps/desktop/src/main/diagnostics/__tests__/hot-cpu-profile-retention.test.ts apps/desktop/src/main/diagnostics/__tests__/hot-cpu-profile-session.test.ts apps/desktop/src/renderer/src/features/settings/pages/__tests__/DeveloperPage.test.tsx apps/desktop/src/renderer/src/features/library/__tests__/HotCpuProfileBanner.test.tsx — 28 passing
• pnpm typecheck
• pnpm test — 259 files, 3060 passing, 1 skipped
• pnpm build
• pnpm lint

Built-app smoke from the first commit:

• Launched apps/desktop from the built out/ files through Electron with PWRSNAP_E2E=1 and a temporary PWRSNAP_USER_DATA.
• Armed hot CPU profiling with a low threshold and heap snapshot capture.
• Verified the temporary diagnostics root contained session.json, samples.ndjson, events.ndjson, renderer-hot-0001.cpuprofile, and renderer-hot-0001-start.heapsnapshot.
• Parsed the .cpuprofile JSON and verified it had non-empty nodes and samples arrays.

Review:

• Ran compound-engineering:ce-code-review; it found cleanup/root/race hardening gaps. Fixed them in fix(desktop): harden hot CPU diagnostics cleanup and reran the full local gates above.

Post-Deploy Monitoring & Validation

• Log searches: hot CPU diagnostics, retention-pruned, hot_cpu_diagnostics_reveal_failed, SESSION_CREATE_FAILED, diagnostics:clearHotCpuSessions, handler_threw diagnostics.
• Metrics/signals: developer reports include complete profile handoff text; diagnostics directories contain session.json, samples.ndjson, events.ndjson, .cpuprofile, and optional .heapsnapshot; repeated sessions stay bounded to the latest retained set.
• Healthy signal: a newly captured hot CPU profile can be revealed from the banner, opened in Chrome DevTools, and correlated with sidecar samples/events.
• Failure signal: reveal commands return validation errors for valid sessions, active captures fail to write artifacts, cleanup removes active/non-session directories, or diagnostics storage grows without pruning.
• Rollback/mitigation: disable hot CPU profiling from Settings or unset PWRSNAP_HOT_CPU_PROFILING; if artifact creation/regression is confirmed, revert this PR before the next desktop release.
• Validation window and owner: validate during the first dogfood session after merge; owner is the PwrSnap desktop maintainer running the diagnostics capture.

Checklist

• PR title follows type(scope): description
• Tests added or updated
• Relevant checks pass locally (pnpm test, pnpm typecheck)
• THIRD_PARTY_LICENSES unchanged; no dependency changes
• Screenshot or recording attached for user-facing changes

---