fix: skip calculating per-task memory limit when in off-heap mode (#2462)

Author: andygrove

spark/src/main/scala/org/apache/comet/CometExecIterator.scala

@@ -93,6 +93,14 @@ class CometExecIterator(
     }
     val protobufSparkConfigs = builder.build().toByteArray
 
+    val memoryLimitPerTask = if (offHeapMode) {
+      // this per-task limit is not used in native code when using unified memory
+      // so we can skip calculating it and avoid logging irrelevant information
+      0
+    } else {
+      getMemoryLimitPerTask(conf)
+    }
+
     nativeLib.createPlan(
       id,
       cometBatchIterators,
@@ -107,7 +115,7 @@ class CometExecIterator(
       offHeapMode,
       memoryPoolType = COMET_EXEC_MEMORY_POOL_TYPE.get(),
       memoryLimit,
-      memoryLimitPerTask = getMemoryLimitPerTask(conf),
+      memoryLimitPerTask,
       taskAttemptId = TaskContext.get().taskAttemptId,
       debug = COMET_DEBUG_ENABLED.get(),
       explain = COMET_EXPLAIN_NATIVE_ENABLED.get(),