backlog: close #1222 -- the fix shipped in ddee5c1d and the banner outlived it

[wshallwshall]

#1222's banner still parsed OPEN while its fix was already on main. That is a status field outliving the code -- the same shape as #1106 -- and it costs a dispatched lane: an incoming session reading the ledger would conclude main is still red on the recursion test and could rebuild a shipped fix.

Verified, not assumed

A commit referencing an item is not a discharge of it, so the check was against the item's own terms:

• ddee5c1d is an ancestor of origin/main
• the replacement test is present on main
• the item's own proof condition holds. #1222 demands that the replacement still FAIL if decode_frame's except RecursionError is narrowed to except ValueError. The shipped test injects a RecursionError and asserts SandboxCodecError, so narrowing lets it escape as RecursionError and the test reddens.

The original trigger was a real depth-100000 nesting, which made the test measure the runner rather than the contract -- environment-dependent by a factor of at least six across images. It was fixed by removing the dependence, not by tuning around it; the depth was never raised.

Closed in place, not archived

Convention measured rather than guessed: origin/main's live ledger already carries 69 items with a closed banner, so archival is a separate sweep.

The original filed blockquote is preserved below the new banner rather than deleted. It carries the environment-dependence measurement and the explicit do not fix this by raising the depth ruling, both of which stay useful after the close.

#1152 was checked in the same pass and deliberately stays OPEN

Its open banner is a ruling, not staleness: Identity.allowed_channels is still None, the item's body requires both halves, the ASVS scorecard carrying the 8.2.2 verdict is outside this repo, and ADR 0134 Amendment A records that only the second clause of release exit criterion 12 is satisfied.

Worth stating because the two look identical from the ledger: a stale banner and a deliberately open one are indistinguishable from status alone. Both parsed open, both had a fix land, and only one was wrong. The discriminator is the item's own proof condition.

Verification

Checked with the parser, never a hand scan, since the banner alphabet is machine-parsed and a hand-rolled scan is a second definition of it:

scripts/docs/backlog_status_check.py
OK, 496 backlog items, each declaring exactly one status
docs/BACKLOG.md 260, docs/archive/backlog/BACKLOG-CLOSED.md 236

Found by the backlog-clearing session while refreshing its handoffs; verified, closed and opened by the coordinator.