Implement pinned drop sugar

[P8L1]

Implements fn drop(&pin mut self) as sugar for Drop::pin_drop.

The resolver recognizes the sugar only for the actual #[lang = "drop"] trait, maps accepted impl items to effective pin_drop resolution, and passes marked impl item IDs to lowering. Lowering then emits HIR with ident pin_drop, so existing type checking, Drop validation, drop glue, and direct-call checks remain unchanged.

Drop identity is handled without a hardcoded core::ops::drop::Drop path check. Local #[lang = "drop"] traits are recorded from raw AST attributes, and external Drop identity is checked through external lang-item metadata via defined_lang_items. This avoids calling the all-crates tcx.lang_items() query from resolver.

This intentionally preserves the base pinned-drop behavior where #[pin_v2] types must use pin_drop, while non-#[pin_v2] types may still implement pin_drop.

r? @petrochenkov

Related

#144537
#130494

[P8L1]

Addressed the review feedback and pushed a follow-up CI fix.

Summary of the review-response changes:

• Removed the simple Param::is_pinned_mut_self_receiver helper and inlined the check at its single use site.
• Removed the early local_lang_drop_traits state entirely, so resolution no longer depends on seeing a local #[lang = "drop"] trait before the impl in source order.
• The resolver now detects fn drop(&pin mut self) syntactically in trait impls and performs the trait-item lookup using pin_drop.
• The accepted sugar is validated after resolution against the actual lang Drop trait, so custom traits with drop or pin_drop do not accidentally accept the sugar.
• Kept the HIR ident rewrite only after validation because associated item collection still derives the impl item name from the HIR ident; accepted sugar needs to be collected as pin_drop, not drop.
• Kept the no-core test custom instead of using add-minicore, because minicore already provides #[lang = "drop"] and does not provide the needed Drop::pin_drop setup for this regression.
• Added/updated regression coverage for the source-order case and custom-trait rejection.

I also fixed the latest CI failure in tests/ui/hygiene/assoc_item_ctxt.rs. The issue was that my change narrowed the general E0407 missing-trait-item suggestion path too much, which broke the intentional same-name suggestion case used by the macro hygiene test. I restored the normal E0407 candidate lookup path, so ordinary diagnostics keep the existing multipart span/suggestion behavior, while invalid fn drop(&pin mut self) sugar on custom traits is still rejected with the targeted diagnostic.

I also checked the tcx.lang_items() use in AST lowering. It is phase-safe here: lower_to_hir explicitly ensures get_lang_items(()) before stealing resolver_for_lowering, and get_lang_items collects local lang items from AST/resolver state rather than from HIR. So this access does not recursively force hir_crate, hir_owner, or HIR map construction.

[rustbot]

This PR was rebased onto a different main commit. Here's a range-diff highlighting what actually changed.

Rebasing is a normal part of keeping PRs up to date, so no action is needed—this note is just to help reviewers.

[rust-log-analyzer]

The job pr-check-2 failed! Check out the build log: (web) (plain enhanced) (plain)

Click to see the possible cause of the failure (guessed by this bot)

    Checking rustc_public v0.1.0-preview (/checkout/compiler/rustc_public)
error[E0658]: box pattern syntax is experimental
    --> compiler/rustc_resolve/src/late.rs:3571:31
     |
3571 |         let AssocItemKind::Fn(box Fn { sig, .. }) = kind else {
     |                               ^^^^^^^^^^^^^^^^^^
     |
     = note: see issue #29641 <https://github.com/rust-lang/rust/issues/29641> for more information
     = help: add `#![feature(box_patterns)]` to the crate attributes to enable
     = note: this compiler was built on 2026-04-13; consider upgrading it if it is out of date