fix: unused struct tuple fields on DragonFlyBSD

[SteveLauC]

What does this PR do

Test the DragonFlyBSD CI failure in #2277, theoretically, I should be able to reproduce it in this PR.

Fix that CI failure by converting the type SendfileHeaderTrailer into a real struct and explicitly mark the fields that are not in use as unused with a _ prefix.

Checklist:

• I have read CONTRIBUTING.md
• I have written necessary tests and rustdoc comments
• A change log has been added if this PR modifies nix's API

[SteveLauC]

Maybe this is not a bug but a feature, such a check was introduced in rust#118297: Merge unused_tuple_struct_fields into dead_code

[asomers]

It looks like those fields only exist in order to keep a slice from dropping while we have a pointer to it. What about changing the function signature so we keep the user's original slice around? Something like:

fn send_file_header_trailer(headers: &Option<Vec<IoSlice<'a>>>, trailers: ...) -> Option<libc::sf_hdtr> {
    ...
}

[asomers]

It looks like those fields only exist in order to keep a slice from dropping while we have a pointer to it. What about changing the function signature so we keep the user's original slice around? Something like:

fn send_file_header_trailer(headers: &Option<Vec<IoSlice<'a>>>, trailers: ...) -> Option<libc::sf_hdtr> {
    ...
}

Ahh, I see now why it was done this way. Because sf_hdtr needs its headers and trailers to be in the form of a slice of iovec, which has a different layout than &[&[u8]].

[asomers]

LGTM, except that the commit message is inaccurate. It applies to FreeBSD and Apple as well. But we don't test those on CI with a nightly compiler, which is why we only noticed for DragonflyBSD.