-Winfer-any-kinded opt-in to avoid false positives

[som-snytt]

Fixes scala/bug#13128

[SethTisue]

@som-snytt how does this relate to @eed3si9n's #11157?

[som-snytt]

@SethTisue it means I'm too lazy to follow the developments. If LR likes the tweaked version, then this chicken version is obviated.

[eed3si9n]

@SethTisue To elaborate, there are a few flavors of cases that Scala 2.13.17 kind-checking lint would warn.

1. curried invariant parameter F[Int] def f[F[_]](i: Int)(ev: F[Int]): Int
2. unused type paramater def f1[F[_]](i: Int): Int
3. curried parameter with F[Int] in contravariant position def f2[F[_]](i: Int)(ev: F[Int] => Int): Int

This PR basically backs out all three kind-checking behind the flag. On the other hand, #11157 backs out the warning for case 1, but not case 2 and 3, which is what Lukas demonstrated using the partest. If mine lands, and we sort of agree that case 2 and 3 are legit, then #11157 would supersede this PR?

[lrytz]

#11157 fixes clear false negatives and I think it will be merged. But we might still demote the flag out of lint, to be decided (see this comment: scala/bug#13128 (comment)).