macros: Improve IDE support

[nurmohammed840]

This refectory improve performance and enhance better compatibility with IDEs.

Here is two examples,

Before: Note that there is no inlay hints for function arguments.

After: Fixed inlay hint hints problem. channel(buffer: .., tokio::spawn(future: ..

---

Before: On syntax error. Run Test | Debug button dissipated. which is frustrating because the editor toggles back and forth while typing.

After: It fix that.

---

It introduce two changes:

1. Remove spaned, Because as it's not meant to be used here.
2. The function body is preserved as is.

[nurmohammed840]

I will send a fix tomorrow.

[nurmohammed840]

What’s really blocking the CI is that,

Current changes preserved async body as-is:

#[tokio::main]
async fn test_with_semicolon_without_return_type() {
    #![deny(clippy::semicolon_if_nothing_returned)]
    #[deny(clippy::semicolon_if_nothing_returned)]
    0;
}

So would convert into:

fn test_with_semicolon_without_return_type() {
    let body = async {
        #![deny(clippy::semicolon_if_nothing_returned)] // ❌
        #[deny(clippy::semicolon_if_nothing_returned)]
        0;
    };
    ...
}

I didn’t realize this, The previous implementation moved global inner attributes outside:

#[deny(clippy::semicolon_if_nothing_returned)]
fn test_with_semicolon_without_return_type() {
    let body = async {
        #[deny(clippy::semicolon_if_nothing_returned)]
        0;
    };
    ...
}

There are many ways to address this issue.
@Darksonn, The real question is, is it worth this fix? The documentation doesn’t mention this behavior.

[Darksonn]

If you believe the output changes are acceptable, you can update the outputs to make the tests pass. See tests-build/README.md for instructions. Then we will review whether the changes are important.

But keep in mind that just because something is not explained in the documentation, that doesn't mean we can just change it. We don't want to break existing users of the macro.

[davidbarsky]

Speaking as a person who works on rust-analyzer:

• I think the span changes are probably good?
• From a hygiene perspective, I worry about the type inference breakage. I think it's important to those to avoid breaking users.
• From a general "how does this this macro work does it work for IDEs" perspective, having the macro return the original tokens such that the expanded (but incorrect!) macro looks mostly correct would go a long way in terms of providing a good IDE experience.

[nurmohammed840]

Oh! It turns out span is used for type suggestion in return or end statements, so I'll revert the previous change.

Seems like it also fix some code suggestion: #3766