chore: 添加完整的诊断输出以追踪 NPE 根本原因

在关键调用链的每个层级添加 System.err 诊断输出：

1. Fork.java:96 - processor 调用前后
   - 记录 input.getData(), acc, branchCount 等关键状态

2. AiStart.java:605 - merge 调用前后
   - 记录 acc, data 参数和 merge 结果

3. Tip.merge() - merge 内部
   - 记录 this, other 参数和 null 处理逻辑

这将帮助我们理解：
- 哪个分支的数据为 null
- null 是在哪个环节产生的
- Fork 的聚合逻辑是如何执行的
- 完整的数据流动路径

相关 issue: #247

Author: CodeCasterX

framework/fel/java/fel-core/src/main/java/modelengine/fel/core/util/Tip.java

@@ -118,12 +118,31 @@ public Tip addAll(Map<String, Content> args) {
      * @return 表示当前的 {@link Tip}。
      */
     public Tip merge(Tip other) {
+        // === DIAGNOSTIC #5: Tip.merge() 开始 ===
+        System.err.println(String.format(
+            "[DIAG-Tip.merge-START] thread=%s, this=%s, other=%s, other_is_null=%b",
+            Thread.currentThread().getName(), this, other, (other == null)
+        ));
+
         // 防御性处理：在并发场景下，Fork.join() 可能传入 null
         // 参考：https://github.com/ModelEngine-Group/fit-framework/issues/247
         if (other == null) {
+            System.err.println(String.format(
+                "[DIAG-Tip.merge-NULL] thread=%s, other is null, returning this=%s",
+                Thread.currentThread().getName(), this
+            ));
             return this;
         }
-        return this.addAll(other.values);
+
+        Tip result = this.addAll(other.values);
+
+        // === DIAGNOSTIC #6: Tip.merge() 结束 ===
+        System.err.println(String.format(
+            "[DIAG-Tip.merge-END] thread=%s, result=%s",
+            Thread.currentThread().getName(), result
+        ));
+
+        return result;
     }
 
     /**

framework/fel/java/fel-flow/src/main/java/modelengine/fel/engine/activities/AiStart.java

@@ -595,14 +595,21 @@ public final AiState<Tip, D, Tip, RF, F> runnableParallel(Pattern<O, Tip>... pat
         }
 
         AiState<Tip, D, Tip, RF, F> state = aiFork.join(Tip::new, (acc, data) -> {
-            // 防御性处理：Fork 的某些分支可能返回 null（特别是并发场景下的竞态条件）
-            // Tip.merge() 会处理 null 情况，参考：https://github.com/ModelEngine-Group/fit-framework/issues/247
-            if (data == null) {
-                log.warn("Fork.join reducer received null data in iteration, this may indicate a race condition. " +
-                        "Tip.merge() will handle this defensively. acc={}, thread={}",
-                        acc, Thread.currentThread().getName());
-            }
-            return acc.merge(data);  // Tip.merge() 内部会处理 null，返回 this
+            // === DIAGNOSTIC #3: AiStart reducer 调用 merge 之前 ===
+            System.err.println(String.format(
+                "[DIAG-AiStart:605-BEFORE] thread=%s, acc=%s, data=%s, data_is_null=%b",
+                Thread.currentThread().getName(), acc, data, (data == null)
+            ));
+
+            Tip mergeResult = acc.merge(data);  // Tip.merge() 内部会处理 null
+
+            // === DIAGNOSTIC #4: AiStart reducer 调用 merge 之后 ===
+            System.err.println(String.format(
+                "[DIAG-AiStart:605-AFTER] thread=%s, mergeResult=%s",
+                Thread.currentThread().getName(), mergeResult
+            ));
+
+            return mergeResult;
         });
         ((Processor<?, ?>) state.publisher()).displayAs("runnableParallel");
         return state;

framework/waterflow/java/waterflow-core/src/main/java/modelengine/fit/waterflow/domain/states/Fork.java

@@ -93,7 +93,24 @@ public synchronized R process(FlowContext<O> input) {
                         acc = Tuple.from((R) "", 0);
                     }
                 }
-                acc = Tuple.from(processor.process(acc.first(), input.getData()), acc.second() + 1);
+
+                // === DIAGNOSTIC #1: Fork.join wrapper 调用 processor 之前 ===
+                Object inputData = input.getData();
+                System.err.println(String.format(
+                    "[DIAG-Fork:96-BEFORE] key=%s, thread=%s, branchCount=%d/%d, acc.first=%s, input.getData=%s, input.getData_is_null=%b",
+                    key, Thread.currentThread().getName(), acc.second(), forkNumber.get(),
+                    acc.first(), inputData, (inputData == null)
+                ));
+
+                R processedResult = processor.process(acc.first(), inputData);
+
+                // === DIAGNOSTIC #2: Fork.join wrapper 调用 processor 之后 ===
+                System.err.println(String.format(
+                    "[DIAG-Fork:96-AFTER] key=%s, thread=%s, processedResult=%s, processedResult_is_null=%b",
+                    key, Thread.currentThread().getName(), processedResult, (processedResult == null)
+                ));
+
+                acc = Tuple.from(processedResult, acc.second() + 1);
                 accs.put(key, acc);
 
                 if (acc.second() == forkNumber.get()) {