fix(RecordAssembler): optimize field creation using Accessor

Author: Gezi-lzq

core/src/main/java/kafka/automq/table/process/RecordAssembler.java

@@ -19,6 +19,7 @@
 
 package kafka.automq.table.process;
 
+import org.apache.avro.util.internal.Accessor;
 import org.apache.kafka.common.cache.LRUCache;
 
 import org.apache.avro.JsonProperties;
@@ -145,7 +146,9 @@ private AssemblerSchema buildFinalAssemblerSchema() {
         List<Schema.Field> finalFields = new ArrayList<>(baseRecord.getSchema().getFields().size() + 3);
         Schema baseSchema = baseRecord.getSchema();
         for (Schema.Field field : baseSchema.getFields()) {
-            finalFields.add(new Schema.Field(field, field.schema()));
+            // Accessor keeps the original Schema instance (preserving logical types) while skipping default-value revalidation.
+            Schema.Field f = Accessor.createField(field.name(), field.schema(), field.doc(), Accessor.defaultValue(field), false, field.order());
+            finalFields.add(f);
         }
 
         int baseFieldCount = baseSchema.getFields().size();