[FIR] Use lower bound of supertype in findClassRepresentation for intersection types

compiler/testData/codegen/box/fir/intersectionWithCapturedTypeWithRawUpperBound.kt

@@ -0,0 +1,43 @@
+// TARGET_BACKEND: JVM
+// WITH_STDLIB
+// MODULE: lib
+// FILE: lib.kt
+interface A<T1> {
+    fun foo()
+}
+
+interface B<T2> : A<T2>
+
+interface C : B<String>
+
+// FILE: FirstHolder.java
+public interface FirstHolder<K1> {
+    K1 getValue();
+}
+// FILE: SecondHolder.java
+public interface SecondHolder<K2 extends A> { // <--- A is raw
+    K2 getValue();
+}
+
+// FILE: HolderImpl.java
+public interface HolderImpl extends FirstHolder<C> {}
+
+// MODULE: main(lib)
+// FILE: main.kt
+fun test(s: Any) {
+    require(s is SecondHolder<*> && s is HolderImpl)
+    /*
+     * - Type of s.getValue() is ft<it(A<kotlin/Any?> & C), C?>
+     * - foo` is found in scope of it(A<kotlin/Any?> & C)
+     *   - there are two candidates (A<Any>.foo and C.foo)
+     *   - they have same signatures, so we chose first of them (`A<Any>.foo` in this case)
+     * - fir2ir needs to find correct f/o lookup tag for this call
+     * - `findClassRepresentation` for `ft<it(A<kotlin/Any?> & C), C?>` in this case should return `null`,
+     *     since there is no actual representation for this type
+     * - if it won't then fir2ir may decide to generate f/o with `C` lookup tag, based on `A.foo` function,
+     *     which is incorrect (there is another f/o in the hierarchy between `C` and `A`
+     */
+    s.getValue().foo()
+}
+
+fun box(): String = "OK"