feat(ui): add native full-width desktop status bar

[matthias-scale]

Intent and provenance

Port RaviTharuma’s native full-width Herdr status row from source head af5ac9d9b1f6425be600cbb904fda905ea882eb9 onto the maintained fork without mutating upstream PR herdrdev#1742.

• Target base: fork/pr-base
• Current head: cf171dcc4a8c19b5b3b45e6ad6c48d0a3e477a66
• Source author: Ravi Tharuma <RaviTharuma@users.noreply.github.com>
• Provenance merge: 9371011c5bc036c1bbeec77416f3e86e551ca58d retains the exact source commit as its second parent.

Binding UI contract

The follow-up contract supersedes the source field set:

• One full-width row remains above the sidebar/tab surface.
• All content is right-aligned; the entire left side stays blank.
• Visible fields, in order: folder → branch → device name → CPU → memory.
• Session identity and workspace/tab/pane numbers remain internal-only.
• Network details, username, Herdr version, battery, date, and time are neither rendered nor collected for this row.
• Optional context elides folder, branch, then device. CPU and memory remain required at ordinary desktop widths and fall back to CPU --% and MEM --/-- GiB.

Acceptance criteria and evidence

• Exact source provenance retained — af5ac9d9b1f6425be600cbb904fda905ea882eb9 is in the branch ancestry with RaviTharuma attribution; this PR targets fork/pr-base.
• One full-width desktop row without geometry regressions — characterization covers sidebar/tab/terminal placement, mobile exclusion, mouse targets, collapsed/compact, multipane, and empty states.
• Folder, branch, device, CPU, and memory only; entirely right-aligned — renderer and negative assertions reject username, network, session identity, and Herdr version.
• CPU/MEM formatting and responsive behavior — 135 focused status tests pass, including stable GiB/% units, unavailable fallbacks, 120-column visibility, and derived required-width elision.
• Bounded non-blocking collection — renderer performs no I/O; sampling uses a two-second cadence, one in-flight sample, four-second staleness, and native/local platform sources with no public-IP request.
• Independent review — Codex and Gemini both report PASS for the final correction; no P0/P1/P2 findings.
• Exact-head cross-platform CI and composition — GitHub Linux/macOS/Windows/Nix/website checks must pass at the current head; after status lands, fork PR feat(ui): focus the sidebar around a spaces tree with per-workspace agent disclosure #1 is merged with the new base and the combined branch receives full checks and independent review.

Verification

• ZIG=/opt/homebrew/opt/zig@0.15/bin/zig just test-one status — 135 passed
• git diff --check — clean
• Codex review — PASS
• Gemini review — PASS
• Full local just check reached a known macOS live-handoff fixture failure already fixed by fork PR feat(ui): focus the sidebar around a spaces tree with per-workspace agent disclosure #1; the final combined branch must pass the complete suite.

Merge boundary

This PR contains teammate-authored work. Merge approval must be pinned to this exact PR head after all checks and review gates are green. Any later push invalidates that approval.

[matthias-scale]

/acode review pass — admin-merging

✅ Auto-merging — 38 files / 3068 LOC, no risky touch, agents clean. Canary catches escapes at 10%.

QA scenarios (post-promote sanity)

• Desktop 200x50: status row occupies row 0 at full client width; left 102 columns blank, content terminates at the final column (right-aligned)
• Field order and contents: folder (…/tmp/herdr-qa-cf171dcc/repo) -> branch (feat/native-status-…) -> device (Matthiass-MBP) -> CPU 65% -> MEM 46.7/64.0 GiB, positions strictly ascending
• Negative field assertions on the rendered row: no username, no Herdr version, no IP/network quad, no battery, no clock time, no date, no workspace/tab/pane numbers
• Live-data provenance: rendered hostname matches hostname -s and MEM total 64.0 GiB matches sysctl hw.memsize; CPU value varied across samples (52% -> 65% -> 41%), proving live sampling rather than a stub
• Format assertions: CPU matches /CPU \d{1,3}%/ and MEM matches /MEM \d+.\d/\d+.\d GiB/
• Responsive width sweep 160/120/100/80/60/40/24: all five fields visible down to 100 cols (120-col visibility confirmed), folder elides first at 80 cols leaving branch+device+CPU+MEM, and the row is excluded entirely at mobile widths (<=60)
• No geometry regression: sidebar and tab surface render immediately below the status row across every tested width
• Unit corroboration: just test-one status -> 135 tests run, 135 passed, matching the PR's stated evidence
• CI at exact head: build, check (ubuntu/macos/windows), Windows ConPTY package, flake check, conventional-commits, check-contributor all pass at cf171dc

Evidence-only advisory (non-blocking)

• Unavailable-metrics fallback (CPU --% / MEM --/-- GiB): The fallback strings could not be forced on a live macOS run because sampling always returned real CPU/memory values. Verified only through the unit suite (135/135 pass), not through a driven session. All other contract fields were asserted against the real rendered row. (P2, not agent-fixable; retained for human follow-up)

---

Resolved — 8/8 CI green · 0/0 threads · agents clean

• CI: build, check-contributor, Windows ConPTY package, check (macos-latest), check (ubuntu-latest), check (windows-latest), conventional-commits, flake check
• Review threads: 0/0 resolved
• Review-pass agents: QA PASS, API contract 0 finding(s), 1 evidence-only advisory/advisories, follow-ups triaged
• Risk classifier: no risky touch (no billing/auth/migrations/RLS/new-UI-flow); 38 files / 3068 LOC (≤10 / ≤1000)

---

Generated by /amerge against cf171dcc. Re-runs on this SHA edit this comment.