test_gap_specabi_reassign: reassigned values read back as 0 (untracked, not in gap_snapshot.json)

[proggeramlug]

test_gap_specabi_reassign diverges from node on a macOS arm64 build of 55fd197d5
(+ PR #7999, which cannot affect program output — it only adds a link-time archive check).

$ diff node.out perry.out
3c3
< plain: 99 101 2
---
> plain: 0 0 2
5c5
< captured: 77:2
---
> captured: 0:2

Reassigned values read back as 0. The shape — a value that was written and then reads as
zero, on the specialized-ABI path — is a codegen/spec-ABI defect, not a runtime or link one.
Compile and run both exit 0; it is an output mismatch, so the harness classifies it
parity_fail.

Why it is being filed rather than triaged onto an existing entry

It is not in test-parity/gap_snapshot.json, so nothing is tracking it and nothing can
go red on it. It was found during a full local gap run for #7629
(PERRY_SKIP_BUILD=1 ./run_parity_tests.sh --filter test_gap_, 538 pass / 15 fail /
1 compile-fail / 0 crashed): 14 of the snapshot's 15 entries reproduced, and this was one of
two failures outside it.

Whether it is macOS-only or the Linux snapshot is stale is unresolved — there is no
test-parity/gap_snapshot.macos.json in the tree, so no macOS gap baseline has ever been
recorded, and the required gap shards run only on ubuntu-latest. The first thing to
establish is which of those two it is; that also decides whether a per-host snapshot is
worth committing.