perf: benchmark startup, focus, tab switching, and multi-pane scaling

[kordless]

Summary

Create repeatable performance baselines for current operating systems and fix regressions revealed by them.

Upstream context:

• Slow tab changes and initial focus on MacOS 26 vercel/hyper#8103
• Performance + Animation Lags on 3.1.0 (stable) vercel/hyper#5754
• Performance Issue At Launching on 3.1.x(stable) Irrelevant to Plugins vercel/hyper#5775
• Why is Hyper startup slow on Windows? vercel/hyper#2980

These reports are demand signals. Hyperia's sidecar, web panes, agent UI, and newer Electron stack require direct measurement.

Metrics

Measure cold and warm launch, tab switch to focused terminal, focus restoration, idle and active CPU and memory, scaling at 1, 10, and 50 panes, high-throughput output, and mixed terminal, web, sticky, and agent layouts.

Acceptance criteria

• A documented benchmark harness produces machine-readable results.
• Cover Windows x64, macOS arm64 and x64 where available, and Linux.
• Record Electron, renderer, OS, architecture, and build mode.
• CI stores benchmark artifacts without using noisy results as a hard gate initially.
• Define budgets for startup, tab switching, idle CPU, and memory.
• Profile and ticket budget violations.
• Add regression tests for concrete fixes delivered through this epic.

Provenance

The immediate motivation is Hyper's current macOS tab and focus complaint, but this ticket measures Hyperia before assuming the same cause.