Ensure ChromeTraceTest event timestamps are emitted in correct invervals

[WojciechMazur]

Fixes #21882 flaky test

[prolativ]

Not sure if this is just a matter of personal taste or if it might make some difference in the actual behaviour:
Why not to compute the expected end time of sleep before the lop and then compare it with System.nanoTime in the loop?

[WojciechMazur]

It might behave differently. If we just sleep again using duration then in the worst case, we can sleep for 1.99x of expected time. To prevent that, we'd need to calculate the remaining time after pre-mature wake up anyway.

1 possible improvement could be potentially usage of TimeUnit.sleep(ofUnit: Long) but that would probably make it more complicate. Previously I though it's available only since JDK 9

[prolativ]

Of course I didn't mean sleeping twice with the same duration. I was thinking of something like

  private def sleep(duration: FiniteDuration): Unit = {
    // A bit of additional precautions to ensure we don't continue execution to early
    // Both LockSuppport and Thread.sleep can return earlier then expected (depending on OS)
    // var remainingNanos = duration.toNanos
    var currentNanos = System.nanoTime()
    val deadline = currentNanos + duration.toNanos
    while
      currentNanos = System.nanoTime()
      currentNanos < deadline
    do
      val remainingNanos = deadline - currentNanos
      val millis = NANOSECONDS.toMillis(remainingNanos)
      Thread.sleep(millis, (remainingNanos % 1.millis.toNanos).toInt)
  }

But maybe this doesn't make that much of a difference