…ed struct

`DefPathData::(ClosureExpr,ImplTrait)` are renamed to match `DefKind::(Closure,OpaqueTy)`.

`DefPathData::ImplTraitAssocTy` is replaced with `DefPathData::TypeNS(kw::Empty)` because both correspond to `DefKind::AssocTy`.
It's possible that introducing `(DefKind,DefPathData)::AssocOpaqueTy` could be a better solution, but that would be a much more invasive change.

Const generic parameters introduced for effects are moved from `DefPathData::TypeNS` to `DefPathData::ValueNS`, because constants are values.

`DefPathData` is no longer passed to `create_def` functions to avoid redundancy.

Now using a window of 3 instead 2 because it removes any
confusion about exactly how consecutive windows overlap

…` format strings

This commit restricts what symbols can be used in a format string for
any option of the `diagnostic::on_unimplemented` attribute. We
previously allowed all the ad-hoc options supported by the internal
`#[rustc_on_unimplemented]` attribute. For the stable attribute we only
want to support generic parameter names and `{Self}` as parameters.  For
any other parameter an warning is emitted and the parameter is replaced
by the literal parameter string, so for example `{integer}` turns into
`{integer}`. This follows the general design of attributes in the
`#[diagnostic]` attribute namespace, that any syntax "error" is treated
as warning and subsequently ignored.

…mented, r=compiler-errors

Restrict what symbols can be used in `#[diagnostic::on_unimplemented]` format strings

This commit restricts what symbols can be used in a format string for any option of the `diagnostic::on_unimplemented` attribute. We previously allowed all the ad-hoc options supported by the internal `#[rustc_on_unimplemented]` attribute. For the stable attribute we only want to support generic parameter names and `{Self}` as parameters.  For any other parameter an warning is emitted and the parameter is replaced by the literal parameter string, so for example `{integer}` turns into `{integer}`. This follows the general design of attributes in the `#[diagnostic]` attribute namespace, that any syntax "error" is treated as warning and subsequently ignored.

r? `@compiler-errors`

…TaKO8Ki

codegen, miri: fix computing the offset of an unsized field in a packed struct

`#[repr(packed)]`  strikes again.

Fixes rust-lang#118537
Fixes rust-lang/miri#3200

`@bjorn3` I assume cranelift needs the same fix.

…iler-errors

more targeted errors when extern types end up in places they should not

Cc rust-lang#115709 -- this does not fix that bug but it makes the panics less obscure and makes it more clear that this is a deeper issue than just a little codegen oversight. (In rust-lang#116115 we decided we'd stick to causing ICEs here for now, rather than nicer errors. We can't currently show any errors pre-mono and probably we don't want post-mono checks when this gets stabilized anyway.)

rustc: Harmonize `DefKind` and `DefPathData`

Follow up to rust-lang#118188.

`DefPathData::(ClosureExpr,ImplTrait)` are renamed to match `DefKind::(Closure,OpaqueTy)`.

`DefPathData::ImplTraitAssocTy` is replaced with `DefPathData::TypeNS(kw::Empty)` because both correspond to `DefKind::AssocTy`.
It's possible that introducing `(DefKind,DefPathData)::AssocOpaqueTy` instead could be a better solution, but that would be a much more invasive change.

Const generic parameters introduced for effects are moved from `DefPathData::TypeNS` to `DefPathData::ValueNS`, because constants are values.

`DefPathData` is no longer passed to `create_def` functions to avoid redundancy.

…mple, r=thomcc

Improve example in `slice::windows()` doc

Fixes rust-lang#118571

Now using a window of 3 instead 2 because it removes any confusion about exactly how consecutive windows overlap