backlog: record why five verified-looking rows must NOT close

[wshallwshall]

Five rows passed a careful verification pass as closeable, and an independent
agent whose only job was to refute them killed all five. Those findings existed
only in a workflow output file in a temp directory. This writes each one into
the row it belongs to.

Nothing closes and nothing opens. Ten insertions, zero deletions, two lines into
each of five rows.

#1061  a merged commit records a deliberate prior refusal
#1226  two limbs; the second is unguarded
#1235  its closing PR falsified the detector's own header on main
#1264  its census keyed on a kind the seat clock never sends
#1332  named by number in #1334 as correctly-open under the
       retired-in-place convention

WHY THIS IS WORTH A CI CYCLE. Without it the next verification pass either
re-derives the same five findings at the same cost, or closes those rows
wrongly -- which is the outcome the refuter just prevented once. #1332 is the
sharpest: closing it would have contradicted a convention the merged ledger
states and names that item by number.

#1235 is not only a blocked closure. Its closing PR left the detector's header
falsified on main, so the merged tree contradicts the reason the row was closed
against. That is a live defect, recorded here because it was found in passing
and would otherwise sit.

Verified structurally rather than by reading the prose. The added lines were
located by POSITION and mapped back to their owning item, because a note in a
row's body need not repeat that row's own number -- matching on the number
found only one of the five:

rows            363 before, 363 after
open rows       248 before, 248 after
status changed  none
diff            +10 / -0, docs/BACKLOG.md only
lines per row   exactly 2 into each of #1061 #1226 #1235 #1264 #1332
all five        still OPEN after

Authored by the Dispatcher seat, who owns ledger prose and ran the refutation.
Verified and landed by the Lander seat.