Use IdentityHashMap for consistent JIT optimization in AbstractCompositeMeter

Replace Collections.emptyMap() with new IdentityHashMap<>() to maintain
consistent map implementation type throughout the object's lifecycle,
enabling JVM scalar replacement optimization and reducing iterator
allocation overhead.

Fixes gh-6811

Signed-off-by: David Mollitor <[email protected]>

Author: belugabehr

micrometer-core/src/main/java/io/micrometer/core/instrument/composite/AbstractCompositeMeter.java

@@ -20,17 +20,22 @@
 import io.micrometer.core.instrument.MeterRegistry;
 import org.jspecify.annotations.Nullable;
 
-import java.util.Collections;
 import java.util.IdentityHashMap;
 import java.util.Iterator;
 import java.util.Map;
 import java.util.concurrent.atomic.AtomicBoolean;
 
 abstract class AbstractCompositeMeter<T extends Meter> extends AbstractMeter implements CompositeMeter {
 
+    @SuppressWarnings("rawtypes")
+    private static final Map EMPTY_CHILDREN = new IdentityHashMap<>(0);
+
     private final AtomicBoolean childrenGuard = new AtomicBoolean();
 
-    private Map<MeterRegistry, T> children = Collections.emptyMap();
+    // IdentityHashMap instead of Collections.emptyMap() keeps implementation type
+    // consistent, which may help JIT optimization.
+    @SuppressWarnings("unchecked")
+    private Map<MeterRegistry, T> children = EMPTY_CHILDREN;
 
     private volatile @Nullable T noopMeter;