[SR-16058][Sema] Consider wrapping type variable layers of optionality when warning about checked casts

[LucianoPAlmeida]

In #39648 I have only considered an extra layer of optionality coming from a value-to-optional constraint restriction.
But that is not the only case where extra optionals need to be added. This can come from overload resolution of a unresolved member producing as result a generic argument that would be bound by context wrapped in layers of optionality. For example the case in the SR:

  let dict: [Int: String?] = [:]
  let foo: Int? = 1
  let _: String? = foo.flatMap { dict[$0] } as? String  // OK

If we look to the constraints debugging we can se that checked cast constraint is
$T3 checked cast to String
Where overload being attempted is
selected overload set choice Int?.flatMap: $T1 == (($T2) throws -> $T3?) throws -> $T3?
So when recording warning fixes we should consider these extra layer of optionality($T3?) in the "from" checked cast sub-expression.

This changes the logic to be based on type of checked cast "from" expression when it is a type variable wrapped in optionals to handle this correctly.

Resolves SR-16058.

[LucianoPAlmeida]

@swift-ci Please smoke test

[xedin]

Sorry it took a bit, I'm going to take a look at this today.

[LucianoPAlmeida]

Sorry it took a bit, I'm going to take a look at this today.

No worries @xedin! Thank you for taking the time for those reviews and for the always insightful feedbacks, really appreciated :)