Create behalf on lingjun.cg@alibaba-inc.com

Run the following benchmark in Linux with 8C32G, it shows poor performance of the new DecimalFormat() when compared to jdk11.
```
@BenchmarkMode(Mode.AverageTime)
@Warmup(iterations = 5, time = 500, timeUnit = TimeUnit.MILLISECONDS)
@Measurement(iterations = 10, time = 500, timeUnit = TimeUnit.MILLISECONDS)
@State(Scope.Thread)
@OutputTimeUnit(TimeUnit.NANOSECONDS)
public class JmhDecimalFormat {

    @Setup(Level.Trial)
    public void setup() {
    }

    @Benchmark
    public void testNewOnly() throws InterruptedException {
        new DecimalFormat("#0.00000");
    }
}
```

the result with jdk 11:
Benchmark Mode Cnt Score Error Units
JmhDecimalFormat.testNewOnly avgt 50 248.300 ? 5.158 ns/op


the result with current jdk:
Benchmark Mode Cnt Score Error Units
JmhDecimalFormat.testNewOnly avgt 50 303.381 ? 5.252 ns/op


When run the other test case to get the flame graph:
```
import java.text.DecimalFormat;
public class DecimalTest {
    private static long start;
    private static long end;
    public static void main(String[] args) {
        start = System.currentTimeMillis();
        int count = Integer.parseInt(args[0]);
        DecimalFormat format;
        for(int i = 0 ; i < count; i++) {
            format = new DecimalFormat("#0.00000");
        }
        end = System.currentTimeMillis();
        System.out.println("Cost " + (end-start) +" ms");
    }
}
```

Compare the flame graph between JDK 11 and JDK 21, it shows "java/util/stream/IntPipeline.findFirst"
task extra 15% CPU time.