Optimize more redundant kotlin/jvm/internal/Refs

The number of initializations of the `value` field before the live range
begins does not really matter so long as we insert a write of a default
value to the local if there were none.

compiler/testData/codegen/bytecodeText/capturedVarsOptimization/capturedInChainOfInlineFuns.kt

@@ -1,6 +1,3 @@
-// IGNORE_BACKEND: JVM_IR
-// TODO KT-36648 Captured variables not optimized in JVM_IR
-
 fun test() {
     var x = 0
     run {

compiler/testData/codegen/bytecodeText/capturedVarsOptimization/capturedInInlineOnly.kt

@@ -1,6 +1,3 @@
-// IGNORE_BACKEND: JVM_IR
-// TODO KT-36648 Captured variables not optimized in JVM_IR
-
 fun test() {
     var x = 0
     run { ++x }

compiler/testData/codegen/bytecodeText/capturedVarsOptimization/capturedValInLambdaInitializedInside.kt

@@ -1,11 +1,4 @@
 // WITH_RUNTIME
-// IGNORE_BACKEND: JVM_IR
-// TODO KT-36648 Captured variables not optimized in JVM_IR
-
-// In JVM IR, SharedVariablesLowering transforms `x` into a shared variable to be able to update it from a lambda,
-// which is a separate function (...$lambda-0).
-// If we keep the existing representation of lambda bodies as separate functions in JVM IR, the only viable option to fix this test
-// seems to support this case in the bytecode optimization pass CapturedVarsOptimizationMethodTransformer.
 
 fun box(): String {
     val x: String

compiler/testData/codegen/bytecodeText/capturedVarsOptimization/capturedVarsOfSize2.kt

@@ -1,6 +1,3 @@
-// IGNORE_BACKEND: JVM_IR
-// TODO KT-36648 Captured variables not optimized in JVM_IR
-
 fun box(): String {
     var xl = 0L     // Long, size 2
     var xi = 0      // Int, size 1

compiler/testData/codegen/bytecodeText/capturedVarsOptimization/withStackNormalization.kt

@@ -1,6 +1,3 @@
-// IGNORE_BACKEND: JVM_IR
-// TODO KT-36648 Captured variables not optimized in JVM_IR
-
 fun add(x: Int, y: Int) = x + y
 
 fun test() {