fix(bench): exempt 3.12.0 No-op rebuild from perf-canary regression gate

CI runner variance on the sub-30ms native no-op rebuild metric caused a
false positive (+396%, threshold 100%) on run 27455727444. None of the
code paths modified by this PR execute on the no-op path — the Rust
pipeline returns at the early-exit branch after Stage 3 before any of
the changed find_enclosing_caller / EDGE_NODE_KIND_FILTER code runs.
Same pattern as 3.11.2:No-op rebuild.

Author: carlos-alm

tests/benchmarks/regression-guard.test.ts

@@ -309,6 +309,18 @@ const SKIP_VERSIONS = new Set(['3.8.0']);
  *   3.11.2:1-file rebuild entry above. Remove once #1440 lands warmups and
  *   3.13+ data confirms the steady state.
  *
+ * - 3.12.0:No-op rebuild — CI runner variance on a sub-30ms native metric.
+ *   The 3.12.0 incremental-benchmark baseline captures noopRebuildMs=23; the
+ *   per-PR perf-canary for PR #1468 (enclosing-caller attribution fix) measured
+ *   dev at 114ms (+396%, threshold 100%) on run 27455727444. None of the code
+ *   paths changed by that PR execute during a no-op rebuild — the Rust pipeline
+ *   returns at the early-exit branch after Stage 3 (detect_changes) with zero
+ *   file changes, before any extraction, edge building, or the modified
+ *   find_enclosing_caller / EDGE_NODE_KIND_FILTER code runs. Root cause is
+ *   shared-runner scheduling jitter amplified by the sub-30ms baseline, identical
+ *   to the 3.11.2:No-op rebuild pattern. Remove once 3.13+ data confirms the
+ *   steady-state.
+ *
  * NOTE: WASM *timing* noise no longer needs per-version entries here — it is
  * handled structurally by WASM_TIMING_THRESHOLD (see above). The 3.11.x
  * entries that remain are kept because they trip the *native* engine too
@@ -330,6 +342,7 @@ const KNOWN_REGRESSIONS = new Set([
   '3.11.2:Full build',
   '3.12.0:Full build',
   '3.12.0:1-file rebuild',
+  '3.12.0:No-op rebuild',
   // tree-sitter-erlang devDependency removed (GHSA-rphw-c8qj-jv84 — malware).
   // The erlang WASM is no longer built, so erlang resolution drops to 0%.
   // These entries exempt the expected precision/recall drop on every build