Should we allow the `case ? =>` syntax for match types?

To summarize the discussion in #12261:

• There are 3 ways to write catch-all patterns for match types (case _ =>, case Any =>, and case ? =>)

• All of them are already in use in the community build (see case ? => in scodecs)

• The current syntax is quite inconsistent:

  type A = _ // NOT OK
  type B = ? // NOT OK
  type M0[X] = X match { case ? => String } // OK
  type M1[X] = X match { case Any => _ }    // OK
  type M2[X] = X match { case Any => ? }    // OK
  val a = "" match { case _: _ => () }      // OK
  val b = try { } catch { case _: ? => () } // OK
  

• Disallowing those standalone ? types would be a relative simple change to the parser (5a81d5aaea3862e58f2c6acbee76c67e8b3cce8b / 8968cc1124b08fb7cf452af0d496e4b8f49a6cfd)

[dwijnand]

So also disallow case _: ? =>, right?

To be honest I find case _: _ => confusingly unnecessary - does it have a use case?

[odersky]

I think we should disallow case ? => . ? should be allowed only as an argument to a type constructor.