ci: offset benchmark cron + scaffold dormant CF Worker fallback

[arimxyer]

Summary

• Move update-benchmarks.yml cron from */30 * * * * to 13,43 * * * * to dodge GitHub's half-hour-boundary throttling. Run history showed ~50% of slots dropped and overnight gaps of 3–4 hours.
• Scaffold a Cloudflare Worker at infra/benchmark-trigger/ that can invoke workflow_dispatch on a Cloudflare cron as a fallback if the offset isn't enough. Inactive by default — wrangler.toml's [triggers] block is commented out, so nothing fires until someone explicitly activates it.

Why

Public repos on round-boundary crons get deprioritized under load. Recent gap pattern (UTC):

Time window	Typical gap
13:00–22:00	~55–75 min
22:00–07:00	2–4 hours
07:00–13:00	1–2 hours

The offset is a free one-line change that typically recovers a meaningful chunk of dropped slots. The dormant worker is the next escalation lever if the offset alone isn't enough — Cloudflare Workers cron triggers fire reliably (sub-minute precision).

The workflow's git diff --quiet data/benchmarks.json && exit 0 step makes duplicate invocations harmless, so once activated, both can fire concurrently without polluting history.

Plan to evaluate

1. Merge.
2. Wait ~1 week, then re-run gh run list --workflow=update-benchmarks.yml --limit 50 to compare gap distribution.
3. If gaps over 2 hours persist (especially overnight), follow infra/benchmark-trigger/README.md to activate the worker. Otherwise leave dormant.

Test plan

• Worker typechecks (bun run typecheck)
• Worker bundles via wrangler deploy --dry-run
• Verify next scheduled run after merge fires on the new :13 or :43 slot
• Re-evaluate gap pattern after ~1 week to decide on worker activation

🤖 Generated with Claude Code