MSQ: Fix two issues with phase transitions. (#17053)

1) ControllerQueryKernel: Update readyToReadResults to acknowledge that sorting stages can
   go directly from READING_INPUT to RESULTS_READY.

2) WorkerStageKernel: Ignore RESULTS_COMPLETE if work is already finished, which can happen
   if the transition to FINISHED comes early due to a downstream LIMIT.

(cherry picked from commit 654e0b444b)

extensions-core/multi-stage-query/src/main/java/org/apache/druid/msq/kernel/controller/ControllerQueryKernel.java

@@ -692,11 +692,10 @@ public class ControllerQueryKernel
   {
     if (stageOutputChannelModes.get(stageId) == OutputChannelMode.MEMORY) {
       if (getStageDefinition(stageId).doesSortDuringShuffle()) {
-        // Stages that sort during shuffle go through a READING_INPUT phase followed by a POST_READING phase
-        // (once all input is read). These stages start producing output once POST_READING starts.
-        return newPhase == ControllerStagePhase.POST_READING;
+        // Sorting stages start producing output when they finish reading their input.
+        return newPhase.isDoneReadingInput();
       } else {
-        // Can read results immediately.
+        // Non-sorting stages start producing output immediately.
         return newPhase == ControllerStagePhase.NEW;
       }
     } else {

extensions-core/multi-stage-query/src/main/java/org/apache/druid/msq/kernel/worker/WorkerStageKernel.java

@@ -183,6 +183,12 @@ public class WorkerStageKernel
       throw new NullPointerException("resultObject must not be null");
     }
 
+    if (phase.isTerminal()) {
+      // Ignore RESULTS_COMPLETE if work is already finished. This can happen if we transition to FINISHED early
+      // due to a downstream stage including a limit.
+      return;
+    }
+
     transitionTo(WorkerStagePhase.RESULTS_COMPLETE);
     this.resultObject = resultObject;
   }