[eric8810]

Part of #541 (not all of it yet).

• The generated const _/entry/registration now carry the impl block's #[cfg]/#[cfg_attr(..., cfg(...))]. The entry is under the same cfg as its block.
• A type parameter that is not used as a callback parameter (e.g. pub fn make<T: Default>(&self) -> T) is now rejected at the method, as R5 requires.
• &'static T parameters are rejected at the method: decoding owns the value and cannot produce a static borrow.
• self: &Self is accepted as a shared receiver by both the macro and the build.rs declaration generator.

Still open in #541: cfg-disabled methods are still validated; marked() in the generator does not filter impl cfg like marked_types() does; callback return types are not checked as BoxFuture<'static, Result<..>>; a camelCase collision across two blocks is not detected.

End-to-end (crates/rutis-bridge/tests/marked_services.rs): a marked block under #[cfg(any())] and one removed by #[cfg_attr(not(any()), cfg(any()))], each naming a function that does not exist, build and serve nothing; a self: &Self method is served. The generator's test checks such a method is declared with no warning.

Note: rustc removes a cfg-disabled item before it expands the attribute macro on it, so the end-to-end case also builds without the guard; the guard keeps the generated entry under the same cfg as its block regardless.

Tests: cargo test -p rutis-bridge --test marked_services, cargo test -p rutis-macros, cargo test -p rutis-bridge --features exports --lib build::exports, cargo check -p rutis -p rutis-loader, cargo fmt --check.

[eric8810]

审核结论：需要修改。

impl 上的 #[cfg] / #[cfg_attr(.., cfg(..))] 带到生成的 const _ 上（lib.rs:280,365）正确；self: &Self 能服务；新的编译期拒绝（只用于返回类型的类型参数、&'static 参数）也对。与 #534/#531/#522 不冲突。

1. build.rs 声明生成器没跟上 self: &Self（crates/rutis-bridge/src/build/exports.rs:1171）：仍要求 receiver.reference.is_some()，而 syn 只在 &self 简写时设置它。结果这种方法运行时可调用，但 TS 声明里没有，还会给出误导的构建警告。修法：与 shared_self 接受同样的接收者，或从本 PR 去掉这部分。
2. 正文写 “Fixes 0.15 review(macros): impl 级 cfg 未带到生成代码，若干绑定错误不在方法处 #541”，合并会关掉 0.15 review(macros): impl 级 cfg 未带到生成代码，若干绑定错误不在方法处 #541，但只修了第 1 项和第 4 项的一部分。仍未做：#[cfg] 掉的方法仍被检查；exports.rs:1011 的 marked() 仍忽略 impl cfg；回调返回类型未检查为 BoxFuture<'static, Result<..>>；两个块的 camelCase 名字冲突未发现。改为 “Part of 0.15 review(macros): impl 级 cfg 未带到生成代码，若干绑定错误不在方法处 #541”。
3. 可选：现有测试只检查属性存在；可在 crates/rutis-bridge/tests/marked_services.rs 加一个 #[cfg(any())] 的标注 impl 做端到端检查。

[eric8810]

接手完成剩余部分（06f62ffa）：

• 已合入当前 main（fix(service): 标注服务复审的五个问题（#510） #534、fix(node): a related later emit waits behind an earlier held emit #557、fix(service): report the effective name of a type's marked blocks #558 之后）；按“不强推”的要求用 merge 而不是 rebase。
• exports.rs 声明生成器接受 self: &Self（89483617，与宏的 shared_self 同样的接收者）；测试另断言这种方法没有构建警告。
• 端到端：marked_services.rs 加了 #[cfg(any())] 与 cfg_attr 关掉的标注 impl（引用不存在的函数），能编译且不被服务；self: &Self 方法能被调用。
• 发现：rustc 在展开属性宏之前就移除 cfg 为假的条目，所以这个端到端用例去掉 guard 也能编译（本地验证过）。0.15 review(macros): impl 级 cfg 未带到生成代码，若干绑定错误不在方法处 #541 第 1 项描述的失败在当前 rustc 上不会出现；guard 保留，作为生成条目与 block 同 cfg 的保证。
• 之前 runtimes-go / runtimes-bun / ci-ok 的失败来自被后续推送取消的运行（39697581742，所有任务 cancelled），不是 PR 引起；随后的 39697856054 已全部通过。