K2: Apply constraints from forks at UNTIL_FIRST_LAMBDA completion phase

Otherwise, OVERLOAD_RESOLUTION_AMBIGUITY is reported for the calls to
the functions annotated as @OverloadResolutionByLambdaReturnType

^KT-43296 In Progress

compiler/testData/diagnostics/tests/inference/inferenceForkRegression.kt

@@ -0,0 +1,26 @@
+// SKIP_TXT
+// FIR_IDENTICAL
+// FULL_JDK
+// WITH_STDLIB
+
+import java.util.concurrent.ConcurrentHashMap
+
+interface ModificationData
+
+fun main() {
+    val updatedFiles = ConcurrentHashMap<String, ModificationData>()
+
+    // HashMap effectively have two Map supertypes: one like Map<K, V> and one Map<K!, V!> (through AbstractMap)
+    // But here, since `updatedFiles` has only a supertype Map<String!, ModificationData!>
+    // `updatedFilesSnapshot` has two supertypes relevant to Map: Map<String!, ModificationData!>  and one Map<String!, ModificationData!>
+    // Actually, we might collapse them, because they're equivalent, but for now we don't
+    val updatedFilesSnapshot = HashMap(updatedFiles)
+
+    // Thus, it leads to constraint system fork during `flatMap` inference
+    // And during overload resolution between two `flatMap` versions with @OverloadResolutionByLambdaReturnType we have to run lambda analysis
+    // So, we have to apply forks to the system on this completion phase too, but still not for the PARTIAL completion mode
+    // Otherwise, OVERLOAD_RESOLUTION_AMBIGUITY happens
+    updatedFilesSnapshot.flatMap { _ ->
+        listOf("")
+    }
+}