[mirror] Upstream PR: per-repaint CPU fix (sirmalloc#397)

[axisrow]

Mirror/tracker for upstream sirmalloc/ccstatusline#397 — "High CPU usage observed with multiple concurrent Claude Code sessions".

Status in our fork: DONE. Merged as PR #2 (final revision includes adversarial-review fixes).

What landed:

• Terminal-width probe: one ps -o ppid= -o tty= -p PID spawn per ancestor (BSD-safe — separate -o flags; FreeBSD's comma-in-header parsing quirk handled), ancestry coverage restored to generations 1–8 with TTY read before PPID termination.
• https-proxy-agent (and its undici graph) loaded via dynamic import only when a proxy is configured; failure paths verified safe on real Node 14.21.3.

Honest benchmark numbers (independent verification; the initially claimed −24% did not replicate and the PR description was corrected):

• CPU per render: 268.8 → 247.9 ms (−7.8%); p50 283.7 → 239.9 ms.
• Follow-up PR perf(widgets): lazy-load widget editors to keep undici out of the render path #3 (lazy widget editors) removes the remaining eager Yoga/undici startup load: total ≈ −15% vs baseline on Node.

Next step: upstream PR referencing sirmalloc#397 from fork main — pending user go.

[axisrow]

Fork-side work is done: the render-path CPU fix is in fork main (35440e4, docs/performance-397-*). Closing; upstream sirmalloc#397 is tracked on its own.