Perf/faster full tree visit

[asdacap]

• Improved tree visitor core utilization.
• On R9 7950x Improved read throughput by 2.5x. With CPU utilization around 80-90%.
• When combined with Feature/verify trie on state finish #7691 (there is a code cache there), can finish mainnet verify trie in 4 minute 30 second.

Types of changes

What types of changes does your code introduce?

• Optimization

Testing

Requires testing

• Yes

If yes, did you write tests?

• Yes

Notes on testing

• Tested manually on a mainnet backup.

[LukaszRozmej]

Can this be done better with voliatale? @Scooletz

[Scooletz]

It can't easily as you want to undo the previous Decrement in an atomic way. You'll need at least one CAS to get it. The case here though is that on faults, you do two CAS operations. In such scenarios you could

var counter = Volatile.Read(ref _slots)
if (counter <= 1)
{
   return false;
}
if (Interlocked.CompareExchange(ref slots, counter -1, counter) == counter))
{
  return true;
}

// slow path in the contended scenario, with loop retry

The question is how contended it is and how likely it's to hit the condition of returning false.

[LukaszRozmej]

Suggested change

	if (tasks != null && tasks.Count > 0) Task.WaitAll(tasks.ToArray());
	if tasks?.Count > 0) Task.WaitAll(tasks.ToArray());

Also we can use do AsSpan() instead of ToArray in .net 9 (update #7585 when this is merged to master)