docs(test): annotate expected impact for fallback case

Author: Stevengre

tests/integration/programs/static-vtable-nonbuiltin-deref.smir.json.expected.NOTES.md

@@ -0,0 +1,48 @@
+# Expected Impact Notes: static-vtable-nonbuiltin-deref
+
+This file marks the expected-output locations directly affected by the
+`GlobalAlloc::Static` / `GlobalAlloc::VTable` non-`builtin_deref` fallback fix.
+
+## Impact 1: top-level alloc kinds are now present
+- Location in expected:
+  - `tests/integration/programs/static-vtable-nonbuiltin-deref.smir.json.expected:2`
+  - `tests/integration/programs/static-vtable-nonbuiltin-deref.smir.json.expected:5`
+  - `tests/integration/programs/static-vtable-nonbuiltin-deref.smir.json.expected:10`
+- Before (pre-fix behavior):
+  - The run panicked during alloc collection with:
+    `Allocated pointer is not a built-in pointer type`
+    (`src/printer.rs:789` on `origin/master` baseline).
+  - So there was no valid `.smir.json`/`.expected` result for this case.
+- Now:
+  - `allocs` contains both `Static` and `VTable` entries.
+- Why this is reasonable:
+  - The test constant is a trait-object reference (`&dyn Debug`), which has
+    both data provenance (to the static) and metadata provenance (to vtable).
+  - Collecting both alloc kinds reflects actual MIR provenance shape.
+
+## Impact 2: trait-object provenance pair is preserved instead of aborting
+- Location in expected:
+  - `tests/integration/programs/static-vtable-nonbuiltin-deref.smir.json.expected:133`
+  - `tests/integration/programs/static-vtable-nonbuiltin-deref.smir.json.expected:134`
+  - `tests/integration/programs/static-vtable-nonbuiltin-deref.smir.json.expected:140`
+- Before (pre-fix behavior):
+  - Traversal hit the assert path while following non-`builtin_deref` container
+    provenance, aborting before full alloc graph serialization.
+- Now:
+  - The serialized constant keeps two provenance pointers (offsets `0` and `8`)
+    and serialization completes.
+- Why this is reasonable:
+  - Fat-pointer layout for trait objects is `(data_ptr, vtable_ptr)`.
+  - Offsets `0` and `8` are the expected two-pointer slots on 64-bit targets.
+
+## Impact 3: deterministic golden normalization for unstable trait IDs
+- Location in expected pipeline:
+  - `tests/integration/normalise-filter.jq:32`
+- Before:
+  - Trait `def_id` values in this case are unstable across runs/environments,
+    causing golden noise unrelated to the semantic bug.
+- Now:
+  - Normalization drops `def_id` object fields before diffing expected.
+- Why this is reasonable:
+  - The test intent is to validate successful static/vtable alloc recovery, not
+    compiler-internal numeric IDs.