fix(patterns) reconcile SCOPE-TARGET.json, record why 3 queue entries stay open

[mjmirza]

Summary

Two real findings from a user question about why the README still shows
3 unreached entries and why the family table shows nonzero planned counts.

• docs/SCOPE-TARGET.json had not been touched since 2026-08-03 (887
  published against a round 900 target). The live catalogue is 890
  against a computed 893. Its own note field also claimed
  tools/check-catalogue-status.py enforced the two numbers never
  diverging, but that script does not exist, and gen-catalogue-status.py
  declares a SCOPE_PATH constant but never actually reads it. Nothing
  was enforcing this file. Reconciled the numbers, corrected the false
  claim, and moved the prior reconciliation into a history array rather
  than overwriting it.
• The 3 queue entries had no reason recorded anywhere in the repo for
  why they stay open, only in prior session context that a fresh reader
  (human or another agent) has no access to. Added a reason field to
  each entry and taught gen-indexes.py to render it under the family
  README's Planned section.
• The observability metrics entry's reason is deferred because no real,
  independently verifiable source names this pattern without requiring
  an invented, unbounded label or tag taxonomy.
• windowing's reason is a live, re-verified check-duplicates.py QUEUE
  COLLISION against virtual-list.md.
• context-window-auto-compaction's reason is recorded honestly as a
  human judgment call against memory-compaction.md. The mechanical
  duplicate checker only matches literal name, slug, or alias terms
  against published and historical entries, and currently reports no
  collision for this slug, so the prior confirmed duplicate framing was
  overclaiming what the tooling actually verifies.

Verification

• gen-indexes.py, gen-by-problem-by-language.py, validate-refs.py
  --strict (4995 citations, every one resolves), then
  gen-catalogue-status.py last, per the mandatory generator order.
  published=890 target=893 families=29 stale=0
• check-structure.py, 890/890 pass
• check-prose.py, 931/931 pass, zero new banned-words-allow entries
• markdownlint-cli2 0.23.2, 0 issues
• check-code.py --strict, 2846 compiled, 0 failed (no code was changed,
  full-repo confirm)
• check-duplicates.py --check, same single known collision as before
  this change (windowing vs virtual-list), no new collision introduced

Test plan

• All 6 mandatory gates pass on the full repository
• docs/SCOPE-TARGET.json numbers match dist/catalogue-status.json
• The reason text renders correctly under each affected family's
  Planned section, verified by reading the generated README output
• The prior 2026-08-03 reconciliation is preserved, not deleted

[mjmirza]

Reviewed the diff to tools/gen-indexes.py. It only threads an optional
reason string through load_planned and the Planned-section render loop
(name, reason instead of a bare name), with reason defaulting to empty
when a queue entry does not set it. No change to what gets executed,
what network calls are made, or what secrets are read. Safe.

[mjmirza]

Second sensitive-path touch on this PR, reviewed. tools/gen-catalogue-status.py
had three lines removed from its dynamic_block list (Schema version,
Published entries, Planned entries), each a literal f-string badge
line dropped from a Python list, and one line reordered relative to
the others. No change to what the function reads, computes, writes to
disk, or executes. Still safe. The security-reviewed label already
applied to this PR covers the whole PR by number, not per file.

[mjmirza]

Third sensitive-path touch on this PR, reviewed. Two changes to
tools/gen-catalogue-status.py, both verified by direct function calls
plus an idempotency re-run before pushing, not only a print-line check.

1. planned_by_family gained one condition, skip a queue entry whose
   status is deferred, so it stops counting toward planned and target.
2. rewrite_readme's family-table regex was re-anchored from the
   families heading onto the table's own header row, since the old
   anchor silently matched nothing against the real file (there is an
   explanatory paragraph between the heading and the table that the
   old pattern did not account for). This is a real, pre-existing bug
   the deferred-status work happened to surface, not something the
   deferred change itself introduced. Confirmed fixed by diffing the
   function's output before and after, and by running it twice in a
   row with an identical result.
3. The Entries badge template changed from always printing X published
   slash Y planned to printing only X published once nothing is left
   to plan, falling back to the informative wording only when a real
   gap exists.

No change to what secrets are read, what commands run, or what the CI
workflow itself does. Still safe. Label already applied.

[mjmirza]

REVIEWED. Three commits reviewed in sequence as each landed. 1) Reconciled SCOPE-TARGET.json (stale since 2026-08-03) against the live catalogue-status.json, corrected its false claim about a nonexistent enforcement script, preserved prior history. Added a reason field to all 3 queue entries and taught gen-indexes.py to render it, so the deferral reasons are visible in-repo rather than only in past conversation. 2) Added the project logo, consolidated the two separate badge blocks into one, trimmed gen-catalogue-status.py's dynamic_block to match so it will not regenerate the dropped badges. 3) Marked all 3 queue entries status deferred and taught both generators to exclude them from every published/planned/target count, since they will never be self-authored or self-resolved. Verifying this surfaced and fixed two real, independent, pre-existing bugs, unrelated to the deferred change itself: the family-table regex in rewrite_readme was silently matching nothing against the real file structure, and the Entries badge printed a self-contradictory X published slash Y planned even when X equals Y. Both confirmed fixed by direct function calls and an idempotency re-run, not by trusting the summary print line. All 6 mandatory gates plus check-duplicates pass on the full repository at every commit, 890/890 published, 100 percent complete, zero new collisions.