Update approval-voting-regression-bench (paritytech#5504)

The accepted divergence rate of 1/1000 is excessive and leads to false
positives especially after
paritytech#4772 and
paritytech#5042, so let's increase
it to 1/100 since we do have some randomness in the system and there is
no point in being that strict.

Fixes: paritytech#5463

---------

Signed-off-by: Alexandru Gheorghe <[email protected]>

Author: alexggh

polkadot/node/core/approval-voting/benches/approval-voting-regression-bench.rs

@@ -74,11 +74,12 @@ fn main() -> Result<(), String> {
 	.map_err(|e| e.to_string())?;
 	println!("{}", average_usage);
 
-	// We expect no variance for received and sent
-	// but use 0.001 because we operate with floats
+	// We expect some small variance for received and sent because the
+	// test messages are generated at every benchmark run and they contain
+	// random data so use 0.01 as the accepted variance.
 	messages.extend(average_usage.check_network_usage(&[
-		("Received from peers", 52941.6071, 0.001),
-		("Sent to peers", 63810.1859, 0.001),
+		("Received from peers", 52941.6071, 0.01),
+		("Sent to peers", 63995.2200, 0.01),
 	]));
 	messages.extend(average_usage.check_cpu_usage(&[
 		("approval-distribution", 6.3912, 0.1),