[JVM_IR] Rebase test of string concatenation codegen.

The JVM IR backend code seems saner to me. The string concatenation
lowering for JVM IR calls the stringPlus intrinsic if there are
only two arguments. That leads to a lot less code:

```
load string
load argument
box argument
call Intrinsics.stringPlus
```

instead of

```
allocate StringBuilder
call StringBuilder.<init>
load string
call StringBuilder.append
load argument
call StringBuilder.append
call StringBuilder.toString
```

This will lead to more boxing, but a lot smaller code. We still
use StringBuilders in JVM IR if there are more than two strings
being concatenated.

compiler/testData/codegen/bytecodeText/boxingOptimization/maxMinBy.kt

@@ -1,6 +1,3 @@
-// IGNORE_BACKEND: JVM_IR
-// TODO KT-36645 Don't box primitive values in string template and string concatenation in JVM_IR
-
 // FILE: list.kt
 
 val intList = listOf(1, 2, 3)
@@ -25,8 +22,6 @@ fun box(): String {
 }
 
 // @BoxKt.class:
-// -- no boxing
-// 0 valueOf
 // -- no compareTo
 // 0 compareTo
 // -- comparisons are properly fused with conditional jumps
@@ -37,3 +32,20 @@ fun box(): String {
 // 4 LCMP
 // 1 IFGE
 // 1 IFLE
+
+// JVM_TEMPLATES
+// -- no boxing but lots of StringBuilder calls
+// 0 valueOf
+// 0 Intrinsics.stringPlus
+// 4 StringBuilder.<init>
+// 8 StringBuilder.append
+// 4 StringBuilder.toString
+
+// JVM_IR_TEMPLATES
+// -- perform boxing and call Intrinsics.stringPlus instead
+// -- of having inlined string builder allocation, appends, and toString
+// 4 valueOf
+// 4 Intrinsics.stringPlus
+// 0 StringBuilder.<init>
+// 0 StringBuilder.append
+// 0 StringBuilder.toString