fix(bench): discard warmup runs in query benchmark median (#1077)

* fix(bench): discard warmup runs in query benchmark median (#1076)

The native fn_deps NAPI path pays a cold-start cost on the first 2-3
calls per worker process — rusqlite statement-cache warmup, OS page
cache fill for the DB file, and NAPI-side static init from
tree-sitter 0.25's transitive crates (indexmap, hashbrown, equivalent)
linked into the .node binary at #1054.

The benchmark's median(5) sample then lands inside cold-start
territory rather than warm steady-state, which is what the regression
gate is supposed to track. On Linux x86_64 CI that surfaced as
fnDeps depth 1 native: 28.7 → 48.6 (+69%) once #1054 enlarged the
binary's init footprint, even though warm per-call latency is
unchanged from v3.9.6.

Run two warmup iterations per (depth, fn) before timing so all 5
sampled timings are warm. Verified on Linux x86_64 (Docker, qemu):
native d1 178.2ms → 119.6ms, now matching the warm d3/d5 plateau
(122/113ms). Wasm path is unaffected by the change since wasm only
ever paid cold-start on its first iteration.

Closes #1076

* fix(bench): bump WARMUP_RUNS from 2 to 3 to match documented worst-case cold-start (#1077)

Author: carlos-alm

scripts/query-benchmark.ts

@@ -108,6 +108,15 @@ console.log = (...args) => console.error(...args);
 
 const RUNS = 5;
 
+// First 2-3 native fnDeps calls per process pay a cold-start cost (rusqlite
+// statement-cache warmup, OS page cache for the DB file, NAPI-side static
+// init from tree-sitter's transitive crates linked into the .node binary).
+// On Linux x86_64 CI, that pulled median(5) into cold-start territory once
+// tree-sitter 0.25 grew the binary's init footprint (#1076), even though
+// steady-state per-call latency is unchanged. Discard the first WARMUP_RUNS
+// before timing so the metric reflects warm-call latency, not cold-start.
+const WARMUP_RUNS = 3;
+
 function median(arr) {
 	const sorted = [...arr].sort((a, b) => a - b);
 	const mid = Math.floor(sorted.length / 2);
@@ -172,6 +181,9 @@ function selectTargets() {
 function benchDepths(fn, name, depths) {
 	const result = {};
 	for (const depth of depths) {
+		for (let i = 0; i < WARMUP_RUNS; i++) {
+			fn(name, dbPath, { depth, noTests: true });
+		}
 		const timings = [];
 		for (let i = 0; i < RUNS; i++) {
 			const start = performance.now();