Incomplete analysis results are available for light class generation

If we have the following classes

 class A // Kotlin
 class B extends A {} // Java
 class C : B() // Kotlin

 The analysis runs into infinite recursion, because C needs all members of B (to compute overrides),
 and B needs all members of A, and A is not available from KotlinCacheManager.getDeclarationsFromProject() -- it is being computed right now,
 so the analysis runs again...

 Our workaround is:
  * when analysis starts put its trace into user data on the project
  * when a light class is about to generate a stub, it requests a trace, and if there is such an incomplete trace, it is returned
  * the logic of TopDownAnalyzer that works in descending order (down the class hierarchy) guarantees that all the superclasses are already processed

idea/testData/hierarchy/class/type/KJKHierarchy/1_kotlin.kt

@@ -0,0 +1 @@
+public open class A<caret>

idea/testData/hierarchy/class/type/KJKHierarchy/2_MyClass.java

@@ -0,0 +1,3 @@
+public class MyClass<caret> extends A {
+
+}

idea/testData/hierarchy/class/type/KJKHierarchy/3_kotlin.kt

@@ -0,0 +1 @@
+public class B: MyClass()

idea/testData/hierarchy/class/type/KJKHierarchy/KJKHierarchy_verification.xml

@@ -0,0 +1,7 @@
+<node text="Object (java.lang)">
+  <node text="A ()" base="true">
+    <node text="MyClass ()">
+      <node text="B ()"/>
+    </node>
+  </node>
+</node>