Prewarm Ghostty terminals immediately

[almogdepaz]

Summary

• prewarm isolated Ghostty terminals immediately instead of after 750ms
• add debug/perf knobs and attribution for hydration, layout-stable, and Ghostty prewarm timing
• extend terminal-load perf reporting and document measured grid/page-load results

Verification

• bun run typecheck
• bun test (1808 pass)
• desktop grid e2e (21 pass)
• local server-only deploy with --broker=no + manual smoke

[almogdepaz]

reviewed the PR diff with gh pr view 186 / gh pr diff 186; no checkout or code changes.

result: no blocking findings.

nonblocking observations:

• public/app.ts:59 promotes GHOSTTY_PREWARM_DELAY_MS from 750ms to 0ms, so the main merge risk is page-load/low-end-device CPU from starting two isolated Ghostty instances immediately. the PR adds page-load perf reporting (scripts/terminal-load-perf.ts:730, scripts/terminal-load-perf.ts:744) and documents measured no-long-task impact, so i don't see this as blocking; just worth watching after merge.
• layout-stable immediate mode stays debug/perf-gated (public/app.ts:1020) and has targeted e2e coverage for cached grid content and ordering (tests/e2e/grid.e2e.ts:609, tests/e2e/session-switch.e2e.ts:557, tests/e2e/session-switch.e2e.ts:598). test coverage looks adequate for the changed behavior.

merge risk: acceptable.