FIR deserializer optimization: avoid calling findKotlinClass twice

This commit influences enum entries deserialization.
In particular, now we don't deserialize enum entry members,
but deserialize its supertype correctly (see test data changes).

compiler/fir/resolve/testData/builtIns/kotlin.txt

@@ -365,13 +365,13 @@ public abstract interface Comparable<in T> : R|kotlin/Any| {
 public final enum class DeprecationLevel : R|kotlin/Enum<kotlin/DeprecationLevel>| {
     private constructor(): R|kotlin/DeprecationLevel|
 
-    public? final enum entry WARNING {
+    public? final enum entry WARNING : R|kotlin/DeprecationLevel| {
     }
 
-    public? final enum entry ERROR {
+    public? final enum entry ERROR : R|kotlin/DeprecationLevel| {
     }
 
-    public? final enum entry HIDDEN {
+    public? final enum entry HIDDEN : R|kotlin/DeprecationLevel| {
     }
 
 }

compiler/fir/resolve/testData/loadCompiledKotlin/annotations/EnumArgumentWithCustomToString.txt

@@ -1,9 +1,7 @@
 public final enum class E : R|kotlin/Enum<test/E>| {
     private constructor(): R|test/E|
 
-    public final enum entry CAKE : R|test/E| {
-        public open fun toString(): R|kotlin/String|
-
+    public? final enum entry CAKE : R|test/E| {
     }
 
 }