fix(gc): pin the two per-thread scanner latches #8552 added (main is red)

changelog.d/8557-holders-scanner-latch-frontier.md

@@ -0,0 +1,15 @@
+### Fixed
+
+- **GC holders census: pin the two per-thread scanner latches that left `main`
+  red.** #8545 taught `scripts/gc_runtime_root_holders.py` to parse
+  `perry_thread_local!` declarations, and #8552 replaced process-global scanner
+  registration latches with per-thread `Cell<bool>` ones. Each change is correct
+  in isolation; together the census gained sight of two latches that no verdict
+  covered (`messaging.rs`'s `GC_SCANNER_REGISTERED`, `native_this_alias.rs`'s
+  `SCANNER_REGISTERED`), and an unclassified holder fails the gate by design.
+
+  Both are idempotence flags for `ensure_*_scanner_registered()` — they store a
+  boolean, never an address — so they are pinned on the identity-ratcheted
+  frontier. That placement is forced rather than chosen: they are rule T (the
+  census cannot see through the type) and the gate rejects a `holders` verdict
+  for a rule-T declaration as stale.

scripts/gc_runtime_root_holders.json

@@ -703,6 +703,10 @@
       "file": "crates/perry-runtime/src/map.rs",
       "name": "TEST_FORCE_HELPER_GC"
     },
+    {
+      "file": "crates/perry-runtime/src/messaging.rs",
+      "name": "GC_SCANNER_REGISTERED"
+    },
     {
       "file": "crates/perry-runtime/src/module_require/path_registry.rs",
       "name": "MODULE_PATH_REGISTRY",
@@ -892,6 +896,10 @@
       "file": "crates/perry-runtime/src/object/native_this_alias.rs",
       "name": "ALIAS_ACTIVE"
     },
+    {
+      "file": "crates/perry-runtime/src/object/native_this_alias.rs",
+      "name": "SCANNER_REGISTERED"
+    },
     {
       "file": "crates/perry-runtime/src/object/prop_plan.rs",
       "name": "READ_PLAN_CACHE"