Fix synthesis of mirrors for GADT with dependent type parameters

[eejbyfeldt]

This extends the current logic in Synthesizer.scala to instead of doing
a single replacement do it until a fixed point is reached.

Fixes #23774

How much have you relied on LLM-based tools in this contribution?

Extensively, for exploring the code base and determining where the issue was.

How was the solution tested?

New automated tests

[bishabosha]

not really sure about these types, is a wildcard more appropriate?

[eejbyfeldt]

not really sure about these types, is a wildcard more appropriate?

But that sounds like a different change? To me it seems this MR is not really changing that logic, just applying the current logic in a way that handles more cases.

You will get similar (at least in my understanding) types if you do

scala> enum E[T] {
     |   case A[T2]() extends E[Int]
     | }
// defined class E
                                                                                               
scala> summon[scala.deriving.Mirror.SumOf[E[Int]]]
val res0:
  scala.deriving.Mirror.Sum{
    type MirroredMonoType = E[Int]; type MirroredType = E[Int]; type MirroredLabel = "E";
      type MirroredElemTypes = E.A[Any] *: EmptyTuple.type;
      type MirroredElemLabels = "A" *: EmptyTuple.type
  } = E$@102fd71d

[bishabosha]

yeah, i guess its fine then, i guess it might be worth having a type class derivation example test case as well.

[eejbyfeldt]

@bishabosha Added a type class derivation example, let me know if you feel is is motivating enough.