fix(bench): add warmup runs to incremental tiers in benchmark.ts (#1487)

* chore: gitignore napi-generated artifacts in crates/codegraph-core

* chore(tests): remove unused biome suppression in visitor.test.ts

* fix(titan-run): sync --start-from enum and phase-timestamp list with actual phases

* fix(hooks): track Bash file modifications via before/after git status diff

Adds snapshot-pre-bash.sh (PreToolUse Bash) + track-bash-writes.sh
(PostToolUse Bash): the pre-hook captures git status --porcelain to a
per-worktree temp file before each Bash call; the post-hook diffs the
before/after state and appends newly modified or created files to
.claude/session-edits.log.

This closes the gap where files written by sed -i, printf redirects,
tee, heredocs, or build tools (Cargo.lock, lockfiles) were never
recorded, causing guard-git.sh to emit false-positive BLOCKED errors.

Closes #1457

* chore(native): remove dead code (unused var, method, variant, fields)

- clojure.rs: annotate lifetime-anchor assignment to silence false-positive
- cfg.rs: remove never-called start_line_of method
- complexity.rs: remove never-constructed NotHandled variant; convert
  irrefutable if-let patterns to plain let destructures
- dataflow.rs: remove never-read callee fields from CallReturn/Destructured
- incremental.rs: remove never-read lang field from CacheEntry

cargo check and cargo clippy both clean after these changes.

* refactor(native): extract emit_pts_alias_edges params into PtsAliasCtx struct

* fix(wasm): sort call targets by confidence before emit to match native engine

* fix(bench): add 2 warmup runs and raise INCREMENTAL_RUNS to 5 for incremental tiers

* fix(hooks): remove node -e and node -p from read-only skip list in snapshot-pre-bash.sh

node -e and node -p can write files via fs.writeFileSync etc., so they
must not be treated as read-only. Without a pre-hook snapshot those writes
are silently untracked in session-edits.log.

* fix(native): drop struct-pattern braces on unit variants in LocalSource match

* fix(hooks): remove echo, printf, awk from snapshot-pre-bash skip list

All three can write files via shell redirection (echo/printf > f, awk > f),
so they must not bypass the pre-hook snapshot. Follows the same reasoning
that removed node -e and node -p in the previous commit.

* fix(bench): exempt 3.12.0:No-op rebuild from regression guard

CI runner variance on the sub-50ms native no-op rebuild metric.
The 3.12.0 baseline captures noopRebuildMs=30 (build) and 23
(incremental); the per-PR gate on run 27457266151 re-measured dev
at 48ms (+60% and +109%) — both exceeding the 50% noisy-metric
threshold. No no-op hot-path code changed in this PR; build-edges.ts
sort only runs when files are actually re-parsed. Same shape and root
cause as 3.11.2:No-op rebuild.

Author: carlos-alm

scripts/benchmark.ts

@@ -90,7 +90,8 @@ try {
 	if (typeof parser.disposeParsers === 'function') disposeParsers = parser.disposeParsers;
 } catch { /* older release — no worker pool to dispose */ }
 
-const INCREMENTAL_RUNS = 3;
+const WARMUP_RUNS = 2;
+const INCREMENTAL_RUNS = 5;
 const QUERY_RUNS = 5;
 const QUERY_WARMUP_RUNS = 3;
 const PROBE_FILE = path.join(root, 'src', 'domain', 'queries.ts');
@@ -154,6 +155,9 @@ const dbSizeBytes = fs.statSync(dbPath).size;
 console.error(`  [${engine}] Benchmarking no-op rebuild...`);
 let noopRebuildMs = null;
 try {
+	for (let i = 0; i < WARMUP_RUNS; i++) {
+		await buildGraph(root, { engine, incremental: true, exclude: BENCH_EXCLUDE });
+	}
 	const noopTimings = [];
 	for (let i = 0; i < INCREMENTAL_RUNS; i++) {
 		const start = performance.now();
@@ -170,6 +174,10 @@ const original = fs.readFileSync(PROBE_FILE, 'utf8');
 let oneFileRebuildMs = null;
 let oneFilePhases = null;
 try {
+	for (let i = 0; i < WARMUP_RUNS; i++) {
+		fs.writeFileSync(PROBE_FILE, original + `\n// warmup-${i}\n`);
+		await buildGraph(root, { engine, incremental: true, exclude: BENCH_EXCLUDE });
+	}
 	const oneFileRuns = [];
 	for (let i = 0; i < INCREMENTAL_RUNS; i++) {
 		fs.writeFileSync(PROBE_FILE, original + `\n// probe-${i}\n`);

tests/benchmarks/regression-guard.test.ts

@@ -258,6 +258,19 @@ const SKIP_VERSIONS = new Set(['3.8.0']);
  *   exemption above (which was a WASM metric; this is native). Exempt this
  *   release; remove once 3.13.0+ data confirms the steady-state.
  *
+ * - 3.12.0:No-op rebuild — CI runner variance on a sub-50ms native metric.
+ *   The 3.12.0 baseline captures noopRebuildMs=30 (build benchmark) and
+ *   noopRebuildMs=23 (incremental benchmark); the per-PR gate re-measures
+ *   dev on a fresh runner and lands at 48ms (+60%) and 48ms (+109%) on run
+ *   27457266151 — both exceed the NOISY_METRIC_THRESHOLD of 50% due to
+ *   sub-50ms variance on shared runners. This PR (#1487) adds warmup runs to
+ *   benchmark.ts on the no-op and 1-file rebuild tiers; on a true no-op
+ *   rebuild no files are re-parsed and build-edges.ts is never reached, so
+ *   none of the code changes in this branch execute on the hot path. The
+ *   delta is entirely shared-runner scheduling noise. Same shape and root
+ *   cause as 3.11.2:No-op rebuild. Exempt this release; remove once
+ *   3.13.0+ data confirms the steady-state.
+ *
  * - 3.12.0:Full build — root-caused residual feature cost of the Phase 8.x
  *   resolution work on the native engine. The v3.12.0 publish gate first
  *   measured 2231 → 3333 (+49%). Local A/B against a v3.11.2 baseline worktree
@@ -309,6 +322,7 @@ const KNOWN_REGRESSIONS = new Set([
   '3.11.2:No-op rebuild',
   '3.11.2:1-file rebuild',
   '3.11.2:Full build',
+  '3.12.0:No-op rebuild',
   '3.12.0:Full build',
   '3.12.0:1-file rebuild',
   // tree-sitter-erlang devDependency removed (GHSA-rphw-c8qj-jv84 — malware).