[Mangling] Add a new mangling for opaque return type to use when mangling an ObjC runtime name

[theblixguy]

The old remangler does not handle opaque result types and can crash when mangling an ObjC runtime name for a class that is in a local context that returns an opaque type, for example:

func opaqueTypeFunc() -> some MyProtocol {
  class Foo: MyProtocol { ... }
  return Foo()
}

var opaqueTypeVar: some MyProtocol {
  class Bar: MyProtocol { ... }
  return Bar()
}

Introduce a new Qu mangling to represent the opaque return type in old mangling.

Resolves SR-13203

[jckarter]

I think we need to implement a real mangling here. The ObjC runtime name can become load-bearing ABI, if the class name ends up being used in a keyed archiver, nib, or other serialized format, and it needs to be unique relative to any other class names that might be instantiated in the process.

[theblixguy]

@jckarter, I have used a new Qu mangling to represent the return type in the old remangler. Does that seem in line with what you had in mind? (Additionally, do we need a new one if we can reuse Qr here)?

Given we don't allow overloading with opaque return types, I think the new mangled name should be completely unique. Is my assumption correct?

[jckarter]

I think that should work.

[theblixguy]

@jckarter Great! Thank you! I'll add this to Mangling.rst as well.

[theblixguy]

@jckarter I have added it to Mangling.rst. Does everything look good to you?

[jckarter]

Looks good to me. Thanks!

[theblixguy]

@swift-ci please test

[theblixguy]

@swift-ci please test

[theblixguy]

@swift-ci please test Linux

[slavapestov]

This needs an executable test in test/Interpreter/ which prints the runtime name. Otherwise we don't know if the mangling works as specified or not.

[theblixguy]

@slavapestov Sure, I have added some new tests here: #33340