feat(rpl): add item-pattern matching and Send/Sync variance predicates

[jellllly420]

Summary

Add a framework-level foundation for Rudra-inspired Send/Sync bug detection using composable predicates over Rust program properties, rather than a dedicated Rudra checker.

• Parse, validate, lower, and document struct/impl item patterns, ADT/access metavariables, wildcard generics/fields/where clauses, impl bindings, and bundle-level predicate guards.
• Match explicit positive unsafe trait impls at HIR/type level using resolved trait identities and explicit paths/imports, including custom traits and local reexports.
• Evaluate Send ownership/bound predicates and Sync resource-access predicates as correlated relational rows. Preserve the access witness so a bound on one method cannot justify a different method.
• Keep upstream's function/ADT MatchSession path alongside the item matcher. Register upstream's has_attr predicate for the stricter where-clause validation and cover the integration with a regression test.
• Add focused parser, metadata, documentation, and UI coverage, including all 12 pinned Rudra Send/Sync fixtures.

The example detection patterns currently live in UI fixtures; this does not enable new default built-in lints. The work is split into eleven commits and rebased onto master at 6fb8514.

Example enabled by this change

pattern send-variance

patt {
    use core::marker::Send;

    send_variance[
        $Wrapper: adt,
        $Parameter: type,
        $MappedType: type,
    ] = {
        struct $Wrapper<..> { .. }

        $marker:
        unsafe impl<..> Send for $Wrapper<..>
        where ..
        {}
    } where {
        has_type_parameters($Wrapper)
        && type_parameter_of($Parameter, $Wrapper)
        && type_parameter_maps_to($Parameter, $MappedType, $marker)
        && owns_type($Wrapper, $Parameter)
        && !is_send_in($MappedType, $marker)
    }
}

The Sync examples separately check exclusive resource access requiring Send and concurrent resource access requiring Sync, correlating each trait query with its own access context.

Current precision / draft scope

The predicate evaluators are initial approximations, not a sound or complete Send/Sync analysis:

• Ownership follows structural by-value fields, with special handling for Box, PhantomData, pointers/references, and unresolved aliases.
• Shared-reference access is signature-based and recognizes direct safe inherent methods. Body-mediated, trait, guard, alias, and specialized access require further work; by-value signatures can over-report when the body never transfers the value.
• Relational evaluation carries unknown/incomplete results, but the current trait-proof backends still map failed rustc proofs to false. Failure to prove a bound is not proof of a concrete non-Send/Sync counterexample.
• Access-specific trait checks currently accept a proof from either the marker-impl environment or the correlated method environment; a fully combined instantiated environment is future work.

Rudra fixture expectations intentionally reflect these predicate semantics, not identical diagnostic output from Rudra.

Validation

On macOS/aarch64 after the rebase (subsequent changes only corrected commit attribution):

• UI tests with embedded and external patterns: 261 passed, 19 ignored, including all 42 Send/Sync cases.
• Workspace unit/integration tests and the new has_attr regression passed.
• cargo clippy --workspace -- -D warnings and Clippy for the new regression test passed.
• cargo fmt --all -- --check, python3 scripts/check_rpl_patterns.py, and git diff --check passed.
• Documentation test invocation completed; the four available doctests are ignored.

Outstanding smoke-test caveat: lintcheck's third-party cc 1.0.67 run was stopped after 5m38s at approximately 100% CPU. A process sample located the time in unchanged upstream MIR interblock dependency-graph construction. That smoke check remains unverified; lintcheck's harness returns success even when a checked crate fails, so its exit status alone is not treated as a clean result. Cache-related failures were also observed during the initial smoke runs.

---

Earlier implementation commits are attributed to GPT-5.6; the rebase compatibility fix is attributed to GPT-6.

Generated by Codex (GPT-6) on behalf of @jellllly420