M5-06: Performance, timeout and response-size acceptance

[pillowtalk-Qy]

Type: Performance Gate Support
Priority: P1
Initial status: Blocked
Owner: QA/Performance
Depends on: integrated main candidate
Blocks: Gate C when hard fail
Writable: benchmark scripts, size guards, docs/performance-report.md

唯一目标

用 production build 测量真实分布并验证 timeout/size 上限，不只展示最佳一次。

Metric	Target	Hard fail
Fixture response p95	<300 ms	>1 s
Live Quote p95	<8 s	>12 s
Live preflight p95	<15 s	>25 s
First interactive	<2 s	>4 s
Credential verify	<100 ms	>500 ms
Export JSON	<2 MB	>5 MB

验收

[ ] 10 Fixture runs
[ ] at least 5 Live attempts
[ ] min/median/p95/max
[ ] timeout deterministic/visible
[ ] response/export hard size tests
[ ] large raw evidence 不冻结 UI
[ ] production build measured
[ ] hard fail -> Bug or NO-GO/approved Scope Cut

验证

pnpm test:performance
pnpm build
pnpm start

Codex 指令

添加最小可重复 benchmark/size guard，测分布和失败，不做 micro-optimization 猜测。
行为修复另开 Bug Issue；报告机器、SHA、样本数和原始结果。

---

[pillowtalk-Qy]

Started from merged main after #102 cleared the production dependency gate.

[pillowtalk-Qy]

Completed via #105. Exact-head and merged-main CI passed; Fixture/local performance passes with the documented Live scope cut.