Measure and cut the GitHub reads idle Passes make #1992

This was generated by AI during a grilling session on #739.

Split out of #739 (decision Q5a). That work makes Passes skip during a Quota wait and keeps a reserve for running Commands; this issue is the other half: spend less of the GitHub quota in the first place.

Problem

Every scheduled Pass reads GitHub before deciding it has nothing to do. With several hosts authenticated as the same GitHub user, all of them draw on one 5000/hr REST quota. On 2026-11-10 the fleet ran roughly 3,000 requests an hour after the cron schedules were already slowed, and ran the quota out twice that afternoon. How many requests an idle Pass of each kind makes has only been estimated from source, never measured.

What a person needs to do first

Measuring needs production data from the hosts (cron logs, gh api rate_limit snapshots over time, or a trace of gh calls per Pass), which an unattended run cannot reach. Once the per-Pass counts are known, shape the reductions: for example, reading once per repository for several Pass kinds, skipping a Sweep that ran recently, or caching negative results for a short time.

Out of scope

• The Quota wait itself (Defer scheduled passes during GitHub quota exhaustion and reduce idle polling #739's Spec).
• GitHub App authentication, which raises the ceiling instead (separate issue).