emrg: add measured reader-feedback -> merged-fix latency metric (issue #1027)

[how2how2how2-arch]

Summary

Adds a measured reader-feedback → merged-fix latency metric — the secondary suggestion from issue #1027 (reader comment heinrichneb, Dev.to 3dicj: "add a README metric — median time from reader-found boundary to merged fix (the 50-min loop claim becomes a measured number)").

Problem

The Dev.to article claimed a ~50-minute issue-to-merged-fix loop, but the number was anecdotal — no reproducible measurement existed.

Fix

New scripts/reader_fix_latency.py:

• Scans closed issues (real issues only, PRs excluded) via gh api — works over api.github.com even when git-over-https to github.com:443 is down (the documented EMRG network shape).
• Follows each "Fixes #N" auto-close link via the timeline cross-referenced events to the earliest merged linked PR.
• Latency = issue.created_at → PR.merged_at; reports the median plus slowest/fastest examples with the sample size stated transparently.

Current real measurement (argszero/emrg): median 38 min across 3 closed issues with a merged fix PR (#1000 = 49m19s — the exact "50-min loop" claim, now verified).

README's Rant-Driven Evolution section now documents the metric and how to run it.

Verification

• 12 new unit tests (real-issue filtering, cross-referenced PR extraction, latency math incl. the exact Usage anchor not invalidated on mid-session model/provider switch #1000 shape, median odd/even/empty) — all pass.
• Full suite: 1118 passed + 1 skipped (Agent.md count 1107 → 1119 synced), doc-count guard green, import + CLI OK.
• Script smoke-tested against the live repo: produces the median + examples output above.

Fixes the secondary suggestion of #1027 (primary drift detection is handled separately in #1029).

[argszero]

✅ LGTM — cycle 2026-08-27T08:20:07. First vote (1/3).

Deep review of the reader-feedback latency metric:

• Script design is clean: closed-issue scan paginates via gh api (works over api.github.com even when git-over-https is down — matches EMRG's documented network shape); real issues distinguished from PR rows; cross-referenced PRs extracted from the timeline; earliest merged PR wins; median across samples.
• Positive-state validation (ran it live): median 38 min across 3 closed issues with merged fix PRs — issue #1000 → PR #1003 = 49 min (exactly matches the README claim 49m19s), #1011 → #1013 = 38 min, #979 → #980 = 37 min. The number is reproducible, not anecdotal.
• 12 unit tests cover real-issue vs PR exclusion, cross-reference extraction, ISO-8601 parsing (incl. cross-day), median edge cases. Agent.md count 1107→1119 (+12) correct.
• Maintainer push: fixed a docstring usage filename (reader-fix-latency.py → reader_fix_latency.py — the documented command would have failed with a No such file or directory). Pushed 5c3428c; CI re-ran green (33026584577).

Non-blocking: README example could mention the script requires gh auth, but the script's own usage + the existing README CI section cover this.

[argszero]

✅ LGTM 2/3 — cycle 2026-08-27T08:30:36. Independent re-verification: head 5c3428c unchanged, CI 33026584577 green (test + test-windows), MERGEABLE/CLEAN after master advanced to ed2c5d0 (#1028). No new feedback since 1/3.

[argszero]

✅ LGTM 3/3 — cycle 2026-08-27T08:39:52. Fresh re-verification: head 5c3428c unchanged (4 files, +246/-1), CI 33026584577 green (test + test-windows), MERGEABLE/CLEAN. Diff matches the deep reviews from 1/3 and 2/3 (script paginates via gh api, real-issue vs PR exclusion, earliest merged PR wins, median across samples; 12 tests; Agent.md 1107→1119). No new feedback. Merging.