ci(release): M164 arm64 判定权交回编译器 + 发布名单守卫（不翻牌）

根因（拿到的真实证据，不是猜的）：v0.2.2（run 42322062879 / job 128248713926）与 v0.2.3
（run 42429017002 / job 128573547829）两份 check-run 注解逐字一致 —— 失败步骤是第 6 步
`Windows ARM64 工具链体检`，第 7 步 `cargo build` 状态 skipped。也就是说 v0.2.1~v0.2.3 五轮
全红都红在一个**编译前的启发式探针**上，真正的编译器/链接器一次都没说话。该探针的结论还与它
自己打印的快照矛盾：`SDK Lib 版本→um 架构: 10.0.22621.0→um:[arm,arm64,x64,x86,]` 明明在，却
同时报 `kernel32=缺失`（同一个 `find -path` 模式在本地合成目录树上命中）。runner 实况是
Windows 11 Arm64 + Visual Studio 2026 镜像，`VC.Tools.ARM64` 与 `Windows11SDK.22621/26100`
都装好了，M142 的 `setup.exe modify` 反倒返回非 0。

改：
- release-build.yml：探针降级为「只取证」——vswhere 组件查询 + 各目录 `.lib` 计数 +
  arm64 `link.exe` 候选，一条 `::error::` 都不写、永不 exit 1；不再自动补装 VS。红绿只由
  cargo/链接/静态 CRT 导入表门禁决定。cargo build 步给 arm64 显式备齐工具链：
  `rustup target add aarch64-pc-windows-msvc`、`vcvarsall.bat arm64` 取 `LIB`/`INCLUDE`、
  vswhere 定位 arm64 `link.exe` 后导出 `CARGO_TARGET_AARCH64_PC_WINDOWS_MSVC_LINKER`
  （顺序不能反：只给链接器不给 LIB ⇒ LNK1181 kernel32.lib）。
- 新增 packaging/check_publish_slots.py（344 行，零依赖）：名单 ⇄ 产物**两个方向**。
  声明了缺 GUI/CLI ⇒ `::error::` + exit 1；有产物没声明 ⇒ `::warning::` 明说未发布（M142 之后
  这个方向是静默的，裁剪步只 echo 一行 drop:）。另有源码级断言（名单每项必须存在且 mode=ci、
  运行期名单必须等于文件里那行）与 8 条 `--self-test` 注入回归。
- release-publish.yml：裁剪前插一步跑那个守卫，原来内联的齐全性循环搬进脚本（一套判据一处实现）；
  重算 SHA256SUMS 的步骤只负责重算。
- release-guard.yml：PR 阶段跑守卫自测 + 源码级断言。
- 文档同步：release.yml / release-publish.yml / release-build.yml 头注释、packaging/README §12、
  §1 矩阵、§3 构建与发布、§9 支持矩阵、§7.5 arm64 取证、release-notes.md、README/README.zh-CN
  的产物矩阵与说明。口径统一为「arm64 构建取证中、未发布」，不写成已支持。

没做的（按 tower 裁决「方案乙」）：`PUBLISH_SLOTS` 本轮**不加回** arm64。arm64 从未产出过 PE、
本机也证不了，写进名单就等于拿 x64 的发布去赌它（M142 移出名单正是为避免这个事故）。下一 tag 的
CI 真出包后翻一行即可，守卫与 SHA256SUMS 自动跟着扩。

验证：workflow YAML 全部可被 PyYAML 解析；所有 `run:` 块过 `bash -n`；探针步在本机（无 VS/SDK）
实跑 exit 0 并打出实况注解；守卫 `--self-test` 8 条全绿、正反两个方向都在合成 dist 上实测过；
check_release_semantics.py（含 --self-mutation）/ release_marker / make_manifest 自检全绿；
cargo metadata 正常。**arm64 产物从未被构建过、也从未在真机跑过**——本机 `rustc --target
aarch64-pc-windows-msvc` 直接撞上没有 MSVC 链接器（git-bash 的 GNU `link` 被当成 link.exe），
`cargo check --workspace --target aarch64-pc-windows-msvc` 也死在 aws-lc-sys 的 cc-rs 阶段
（GNU host 上没有该 target 的 C 编译器）。真实结论只能等下一 tag 的 CI。

README.md

@@ -264,10 +264,14 @@ Grab the file for your platform from
 **This release publishes `windows-x64` and nothing else.** The publish job keeps only the slots named
 in `PUBLISH_SLOTS` (`.github/workflows/release-publish.yml`) and deletes every other artifact before
 it writes `SHA256SUMS.txt`, so a row marked **not yet published** is a naming rule, not a downloadable
-asset. Of those rows `windows-arm64` is the one CI still builds — natively, on GitHub's
-`windows-11-arm` runner, for evidence — but that slot's toolchain check has not gone green yet, so its
-`.exe` stays out of the release. The row stays because the architecture is still supported; it rejoins
-the download list once the build is green.
+asset. A guard (`packaging/check_publish_slots.py`) watches both directions: a slot named in the list
+whose artifacts are missing turns the release red, and artifacts that exist for an unlisted slot get an
+explicit "not published" warning instead of being deleted quietly. Of those rows `windows-arm64` is the
+one CI still builds — natively, on GitHub's `windows-11-arm` runner (Windows 11 Arm64 + Visual Studio
+2026), for evidence. Every ARM64 failure so far came from a pre-build toolchain probe that aborted before
+`cargo build` ran; that probe is now evidence-only, so an upcoming CI run is the first real answer to
+whether the slot can produce a binary. Its `.exe` stays out of the release until then; the row stays
+because the architecture is still supported.
 
 Each UNIX archive contains the executable plus `LICENSE`, `NOTICE` and `README.md`, in one top-level
 directory. Windows ships a bare `.exe` (a single file cannot carry the texts), so those three