Fix sneaky binary incompatibility in Discover.scala

[lihaoyi]

fixes #2749

The basic problem is that the signature of Discover.apply() changed, but it only changed in the argument's generic type parameters, so it was nominally the same type after erasure, and accepted the Map[Class[_], Seq] even though it was expecting a Map[Class[_], (Seq, Seq)], but later on in the code it would blow up with a ClassCastException

This causes problems with Mill plugins containing external modules, which have their own Discover[T] macro pre-expanded and would not get re-compiled with the new 0.11.3 version of Mill.

This PR works around the problem by adding a forwarder Discover.apply(value: Map[Class, Seq]) that does the right thing and expands the Seq into a (Seq, Seq), while adding a dummy paramter to case class Discover to avoid conflicts. A similar forwarder Discover.<init>(value: Map[Class, Seq]) is needed to preserve binary compatibility (I think?)

Tested manually via ./mill -i dev.run example/basic/1-simple-scala --import ivy:io.chris-kipp::mill-github-dependency-graph::0.2.6 io.kipp.mill.github.dependency.graph.Graph/generate, which fails before this PR and passes after

[ckipp01]

From a pure usability perspective this seems to fix the issues I was seeing locally. From a code perspective 😅 I'll just refer to your expertise here, but it still seems that Mima is unhappy.

[lefou]

Can you please ads @deprecate annotations denoting these additions as bincompat shims. It helps with cleaning later.

[lefou]

I applied more tweaks due to binary compatibility requirements against 0.11.3.