FIR DFA: move eq/notEq null-to-type translation to LogicSystem

This makes the `returns() implies` checker slightly cleaner, and also
fixes the case that I've missed where in RHS of `x ?:` type of `x` was
not set to `Nothing?`.

compiler/testData/codegen/bytecodeText/inlineClasses/unboxInlineClassAfterElvis.kt

@@ -1,4 +1,6 @@
 // !LANGUAGE: +InlineClasses
+// IGNORE_BACKEND_K2: JVM_IR
+// FIR_STATUS: `x ?: x!!` assumed to throw if x is null, so only 2 unboxings
 
 // FILE: utils.kt
 
@@ -16,4 +18,4 @@ fun test(x: UInt?, y: UInt) {
 // 3 INVOKEVIRTUAL UInt.unbox
 
 // 0 valueOf
-// 0 intValue
+// 0 intValue