perf(desktop): stop the statusbar re-rendering per streaming token

[OutThisLife]

First fix found with the render-churn counters from #71925. Cuts total renders during multi-tab streaming by 72%.

The bug

The statusbar subscribed to $focusedSessionState — a projection of $sessionStates, which is republished on every message delta — but reads only three fields off it (busy, usage, turnStartedAt).

@nanostores/react's useStore bails out on reference equality, so a republished state object re-ran useStatusbarItems per token, rebuilding all ~9 statusbar item objects. StatusbarItemView wasn't memoized, so one genuinely-changed item (the running timer, ticking every second) re-rendered the entire bar. Each of those items carries a tooltip, which is why TooltipContent was the single worst component in the whole app at 2,048 wasted renders of 2,304.

The statusbar looked innocent — StatusbarSurface is already memo()'d with a comment explaining it's isolated so "any statusbar-only churn re-renders the bar alone." That isolation was real and still didn't help, because the churn was inside the boundary.

The fix

Adds useStoreSelector beside the existing useSessionSlice in lib/use-session-slice.ts — the same narrowing idea that file already documents, for a scalar instead of a keyed array. Subscribe to the store, but bail out unless the selected value actually changes.

Applied to the three fields the statusbar reads, plus memo() on StatusbarItemView. No behavior change; 30 shell tests pass.

Measured

Five concurrent streaming tabs, node scripts/perf/run.mjs render-churn --spawn --tiles 5 --tokens 240:

metric	before	after
total renders	78,385	21,864 (−72%)
wasted renders	10,432	6,121 (−41%)
TooltipContent	2,304 / 2,048 wasted	143 / 127 (−94%)
StatusbarItemView	2,174 / 1,446 wasted, 594ms	0

sidebar_renders stays at 6 with 0 wasted — the sidebar was already clean and this doesn't disturb it.

Note on the selector contract

select must return a primitive or a referentially stable value; returning a fresh object or array defeats the bail-out and reintroduces the churn. That's documented on the function. usage is passed through as an object deliberately — it's replaced wholesale rather than mutated and only changes when the backend reports new usage, so its reference is a valid bail-out key on its own.

Remaining waste (Block/Ct at 880/880 each, 542ms combined) is in the transcript renderer under minified component names — separate PR.

[github-actions]

૮ >ﻌ< ა ci review

_{ran on 41dd895}

ℹ️ Info

Desktop E2E visual evidence · View test artifacts · View job

1 visual diff.

_{inline evidence upload failed.}

Failed to upload diff-665a0833239e-onboarding-overlay-diff.png with gh image (exit code 1): Error uploading /home/runner/work/_temp/e2e-evidence/diff-665a0833239e-onboarding-overlay-diff.png: step 0 (get upload token): uploadToken not found on repo page — do you have write access to NousResearch/hermes-agent? (or, if NousResearch enforces SAML SSO, authorize at https://github.com/orgs/NousResearch/sso)