Rewrote enum lowering to support usage from other modules.

- Split up enum lowering into two stages: descriptors creation and IR creation.
- Added link stage test.
- Enum.valueOf() uses binary search instead of linear.

backend.native/tests/codegen/enum/linkTest_lib.kt

@@ -0,0 +1,7 @@
+package a
+
+enum class A(val x: Int) {
+    Z1(42),
+    Z2(117),
+    Z3(-1)
+}

backend.native/tests/codegen/enum/linkTest_main.kt

@@ -0,0 +1,7 @@
+import a.*
+
+fun main(args: Array<String>) {
+    println(A.Z1.x)
+    println(A.valueOf("Z2").x)
+    println(A.values()[2].x)
+}

backend.native/tests/codegen/enum/valueOf.kt

@@ -1,8 +1,11 @@
 enum class E {
+    E3,
     E1,
     E2
 }
 
 fun main(args: Array<String>) {
     println(E.valueOf("E1").toString())
+    println(E.valueOf("E2").toString())
+    println(E.valueOf("E3").toString())
 }

backend.native/tests/codegen/enum/values.kt

@@ -1,8 +1,11 @@
 enum class E {
+    E3,
     E1,
     E2
 }
 
 fun main(args: Array<String>) {
+    println(E.values()[0].toString())
     println(E.values()[1].toString())
+    println(E.values()[2].toString())
 }