Fix #9482: implement reflect.Manifest summoning

[bishabosha]

This PR implements summoning for reflect.Manifest, reflect.OptManifest and reflect.ClassManifest following the algorithm present in Scala 2:
https://github.com/scala/scala/blob/0c011547b1ccf961ca427c3d3459955e618e93d5/src/compiler/scala/tools/nsc/typechecker/Implicits.scala#L1519

A special case is added in Definitions for reflect.ClassManifest so that is not de-aliased at the time of Implicit search, added in c84f057, so careful review is required here.

fixes #9482

[bishabosha]

So far there is no restriction on summoning - I can think of several options:

• enable it only in source:3.0-migration - I think this could be a problem as Manifests themselves are not deprecated
• forbid it in source:future - by this time the library may have deprecated/dropped them
• enable it under one of the legacy language features?

[smarter]

I won't be available to review this in the near future but my first reaction here is that this is way too much code and complexity for a legacy and known-broken feature, I also don't understand why some of the logic and error messages reference TypeTag since this isn't something we support at all. IMO, it's fine to add support for the simplest usecases where Manifest is used instead of ClassTag, but we shouldn't aim to perfectly emulate what Scala 2 does (because we'll never match it exactly anyway).

[bishabosha]

The special casing for ClassManifest is causing too much trouble,

And I will try to simplify, there is no need to support type parameters as <:< as is deprecated.

I will also remove the special casing for TypeTag

[smarter]

So far there is no restriction on summoning - I can think of several options:

I suggest emitting a deprecation warning encouraging the use of ClassTag and http://dotty.epfl.ch/docs/reference/metaprogramming/toc.html whenever we materialize a Manifest.

[bishabosha]

see #13142