refactor(ast): make AST structs repr(C). (#4614)

Make structs `#[repr(C)]` without reordering them to benchmark the performance implications of it.

This PR adds 136 bytes of padding in total. You can visit the list of these types [here](https://github.com/oxc-project/oxc/pull/4404/files/)(checkout the `assert_layouts.rs` diff).

Update:

Doesn't seem too bad!

![image](https://github.com/user-attachments/assets/05ac9cc8-7cb0-4ed5-9220-44f65c7f5bb9)

* Linter: We can easily get more than 1% performance gain in the linter to even things out
* Prepass: If we fix our issue with Rust `1.80.0` we gain 13% And it should also be possible to find some other areas there to gain back this one percent.

The most notable thing is that `parser` isn't impacted by this, That is our most optimized crate which every percent counts(and it can be hard to gain back any perf regression there).

crates/oxc_ast_macros/src/lib.rs

@@ -17,9 +17,10 @@ fn enum_repr(enum_: &syn::ItemEnum) -> TokenStream2 {
 /// It is also a lightweight macro; All of its computation is cached and
 /// it only applies the following changes without any complex operation:
 ///
+/// * Prepend `#[repr(C)]` to structs
 /// * Prepend `#[repr(C, u8)]` to fieldful enums e.g. `enum E { X: u32, Y: u8 }`
 /// * Prepend `#[repr(u8)]` to unit (fieldless) enums e.g. `enum E { X, Y, Z, }`
-/// * Prepend `#[derive(oxc_ast_macros::Ast)]`
+/// * Prepend `#[derive(oxc_ast_macros::Ast)]` to all structs and enums
 ///
 #[proc_macro_attribute]
 #[allow(clippy::missing_panics_doc)]
@@ -28,8 +29,7 @@ pub fn ast(_args: TokenStream, input: TokenStream) -> TokenStream {
 
     let repr = match input {
         syn::Item::Enum(ref enum_) => enum_repr(enum_),
-        // In future, we'll add `#[repr(C)]` to structs, but at present this is disabled
-        syn::Item::Struct(_) => TokenStream2::default(),
+        syn::Item::Struct(_) => quote!(#[repr(C)]),
 
         _ => {
             unreachable!()