perf(homepage): aggregate strips in SQL

[VrianCao]

Why

Current dominant CPU outlier is the homepage snapshot refresh (fetch /api/v1/internal/refresh/homepage) at ~13–16ms CPU per invocation.

A big part of that path is materializing heartbeat + uptime strip arrays in JS (looping over many rows) after reading check_results and monitor_daily_rollups.

What

• Change computePublicHomepagePayload() to fetch pre-aggregated strips from D1:
  • Heartbeat strip: checked_at[], latency_ms[], status_codes via json_group_array(...) + group_concat(...)
  • Uptime day strip: day_start_at[], downtime_sec[], unknown_sec[], uptime_pct_milli[] via json_group_array(...), plus rollup total sums for uptime_30d
• JS now assigns the arrays directly (with safe JSON parsing) and only appends the optional “today” partial bucket.

No API shape changes.

Where

• apps/worker/src/public/homepage.ts
• apps/worker/test/public-homepage-compute.test.ts

How to verify

• Local:

  • pnpm --filter @uptimer/worker lint
  • pnpm --filter @uptimer/worker typecheck
  • pnpm --filter @uptimer/worker test

• Prod (after deploy):

  • pnpm --filter @uptimer/worker exec wrangler tail uptimer --format json --sampling-rate 0.999
    • Watch cpuTime for fetch /api/v1/internal/refresh/homepage (expected drop)