[ML] Improve reason when autoscaling capacity cannot be computed (#89316) (#89320)

When we cannot compute autoscaling capacity we call
`MlAutoscalingDeciderService.buildDecisionAndRequestRefresh`.
There are 4 callers. 3 of them set their own specific reason before
the call. However, the method overwrites the reason with the generic
`MEMORY_STALE` msg. This commit addresses this issue which should
bubble up more detailed reasons for those edge cases.

Co-authored-by: Elastic Machine <[email protected]>

Author: dimitris-athanasiou

x-pack/plugin/ml/src/main/java/org/elasticsearch/xpack/ml/autoscaling/MlAutoscalingDeciderService.java

@@ -507,7 +507,7 @@ public AutoscalingDeciderResult scale(Settings configuration, AutoscalingDecider
                 "view of job memory is stale given duration [{}]. Not attempting to make scaling decision",
                 mlMemoryTracker.getStalenessDuration()
             );
-            return buildDecisionAndRequestRefresh(reasonBuilder);
+            return buildDecisionAndRequestRefresh(reasonBuilder.setSimpleReason(MEMORY_STALE));
         }
         // We need the current node loads to determine if we need to scale up or down
         List<NodeLoad> nodeLoads = new ArrayList<>(mlNodes.size());
@@ -1163,7 +1163,7 @@ Optional<NativeMemoryCapacity> calculateFutureAvailableCapacity(PersistentTasksC
 
     private AutoscalingDeciderResult buildDecisionAndRequestRefresh(MlScalingReason.Builder reasonBuilder) {
         mlMemoryTracker.asyncRefresh();
-        return new AutoscalingDeciderResult(null, reasonBuilder.setSimpleReason(MEMORY_STALE).build());
+        return new AutoscalingDeciderResult(null, reasonBuilder.build());
     }
 
     private Long getAnalyticsMemoryRequirement(String analyticsId) {