test: fix flaky test-bench-stream

testDeliveryDoesNotConsumeTimeout gave the benchmark a 20ms timeout
while stalling the consumer for 50ms. Only delivery time is credited
back to the deadline, so the 32 samples still had to run within 20ms,
which is about a 10x margin on an idle machine and not enough on a
loaded CI runner.

Scale the timeout and the stall together so the benchmark's own work
gets 500ms of headroom while the consumer still stalls for longer than
the timeout.

Signed-off-by: Matteo Collina <hello@matteocollina.com>

Author: mcollina

test/parallel/test-bench-stream.js

@@ -151,15 +151,21 @@ async function testCancellationCompletesBenchmarks() {
 
 async function testDeliveryDoesNotConsumeTimeout() {
   const runner = createRunner({ yieldBetweenSamples: false });
+  // The timeout only has to cover the benchmark's own work, which is 32 samples
+  // that do nothing but record a fixed value. Keep it generous so that a loaded
+  // machine cannot exhaust it on its own, and keep the consumer stalled for
+  // longer than the timeout so that the benchmark can only complete when the
+  // time spent delivering records is excluded from the timeout.
+  const timeout = common.platformTimeout(500);
   const completion = runner.bench('slow consumer', {
     samples: 32,
-    timeout: common.platformTimeout(20),
+    timeout,
   }, recordSample);
   const stream = runner.run();
   const iterator = stream[Symbol.asyncIterator]();
 
   await iterator.next();
-  await setTimeout(common.platformTimeout(50));
+  await setTimeout(timeout * 2);
   for (;;) {
     const next = await iterator.next();
     if (next.done) break;