Fix: handle stack frames in the correct order

Commit f3c64ee removed `ObjArray` and
replaced its usage with standard JDK classes. In `Interpreter`, in
particular, an `ArrayDeque` is now used to store
`cx.previousInterpreterInvocations`, which is used to generate the
stack frame information. However, there is one place where `toArray`
is done, and the behavior for `ObjArray` and `ArrayDeque` are different
(swapped).
Unfortunately no tests actually ends up exercising this difference due
to the interpreter function peeling optimization done in
#1510.

We have discovered this problem because, in ServiceNow's fork, we
currently need to disable the function peeling optimization.

rhino/src/main/java/org/mozilla/javascript/Interpreter.java

@@ -14,6 +14,7 @@
 import java.util.ArrayDeque;
 import java.util.ArrayList;
 import java.util.Arrays;
+import java.util.Collections;
 import java.util.HashSet;
 import java.util.List;
 import java.util.Objects;
@@ -951,7 +952,10 @@ public void captureStackInfo(RhinoException ex) {
                 --previousCount;
             }
             array = new CallFrame[previousCount + 1];
-            cx.previousInterpreterInvocations.toArray(array);
+
+            ArrayList<Object> tempList = new ArrayList<>(cx.previousInterpreterInvocations);
+            Collections.reverse(tempList);
+            tempList.toArray(array);
         }
         array[array.length - 1] = (CallFrame) cx.lastInterpreterFrame;