Support Result<(), impl Debug> in #[wasm_bindgen(start)]

[daxpedda]

This adds support to #[wasm_bindgen(start)] to support more types closer to what Rusts fn main() supports while not changing the current behavior of throwing JsValues when available.

Looking at Termination, which is what fn main() uses, I added support for Infallible and Result<(), impl Debug>.
I intentionally left ExitCode unaddressed because I don't know exactly what to throw in that case.

I contemplated introducing a new trait, like JsTermination to mirror exactly what Rust does, but that's not really compatible with the current way #[wasm_bindgen(start)] works. We could however introduce it with a new #[wasm_bindgen(main)] if that is desired? See #3252 for a similar proposal.

The way this problem was solved is by using Autoref-based specialization, see this case study by dtolnay, which is sometimes used to solve this kind of problem in macros.

Fixes #3262.

[ranile]

What's the use case for this? I don't really see how impl Debug is better than String. Anything that's impl Debug can be trivially converted to a string so I'm not really sure about adding support for it

[daxpedda]

The use case is cross-platform compatibility. See #3262.

The reason why usually people use fn main() -> Result<(), anyhow::Error>, or similar, is because it can be used to casually return any error type. You can't use fn main() -> Result<(), String> and return any error, you would have to write a .map_err() conversion for every single ?.

I didn't pick Debug randomly, it is the chosen trait by Rust to return from fn main(), see Termination. So all I'm trying is to make fn main() as compatible as possible between wasm-bindgen and Rust.

[daxpedda]

See also #2994 and #2970.

[daxpedda]

See #3252 (comment).

I will introduce a new attribute that is supposed to only be used with fn main().