[CSSimplify] CGFloat-Double: Rank narrowing correctly when result is injected into an optional

If result of `CGFloat` -> `Double` conversion is injected into an optional
it should be ranked based on depth just like when locator is fully simplified.

For example:

```swift
func test(v: CGFloat?) {
   _ = v ?? 2.0 / 3.0
}
```

In this expression division should be performed on `Double` and result
narrowed down (based on the rule that narrowing conversion should always
be delayed) but `Double` -> `CGFloat?` was given an incorrect score and
instead of picking `?? (_: T?, _: T) -> T` overload, the solver would
use `?? (_: T?, _: T?) -> T?`.

(cherry picked from commit cb876cbd9e)

test/Constraints/bidirectional_conversions.swift

@@ -6,7 +6,7 @@ import CoreGraphics
 
 /////////////
 
-struct G<T> {  // expected-note {{arguments to generic parameter 'T' ('CGFloat?' and 'CGFloat') are expected to be equal}}
+struct G<T> {
   var t: T
 }
 
@@ -32,8 +32,7 @@ func foo4(x: (() -> ())?, y: @escaping @convention(block) () -> ()) -> G<() -> (
 
 func foo5(x: CGFloat?, y: Double) -> G<CGFloat> {
   let g = G(t: x ?? y)
-  // FIXME
-  return g  // expected-error {{cannot convert return expression of type 'G<CGFloat?>' to return type 'G<CGFloat>'}}
+  return g
 }
 
 func foo6(x: Double?, y: CGFloat) -> G<Double> {