move `Alias[Ty|Const|Term]Kind` into `Alias[Ty|Const|Term]`, generalize to not require just a `DefId`

[BoxyUwU]

ty::Ty has a Alias(AliasTyKind, DefId, GenericArgs) variant, whereas ty::Const just has Unevaluated(DefId, GenericArgs).

With mGCA we now have more than just anon consts in Unevaluated and have ways of normalizing that are not by invoking CTFE so this naming no longer makes sense. We're also starting to write code that wants to know what kind of const alias its interacting with (e.g. a projection, inherent, free, etc). This is really easy to check with types but with const generics we have to write awkward code matching on DefKinds and parents.

I expect consts will probably want to go one step further than types and have different representations for different kinds of unnormalized consts. E.g.

struct AliasConst(DefId, GenericArgs);

enum AliasConstKind {
    Inherent, Projection, Free, 
}

enum BikeshedConst {
    ConstItem(AliasConstKind, AliasConst),
    Literal(Ty, LitKind, bool),
    // add stuff which was previously in `ty::ExprKind`
}

enum ConstKind {
    ...

    // remove these
    Expr(ty::Expr),
    Unevaluated(ty::UnevaluatedConst),
    // replace with this
    Bikeshed(ty::BikeshedConst),
}

I'm not sure exactly what we should call Bikeshed/BikeshedConst, maybe just Expr works idk :3

This issue has been assigned to @khyperia via this comment.

[lcnr]

more generally, we should move the Alias[Ty|Const|Term]Kind into Alias[Ty|Const|Term] and stop requiring every alias to be identified by just their DefId. This also unblocks a fix for issue 7 of rust-lang/trait-system-refactor-initiative#8 and the stabilization of inherent associated types.