vtables may be include auto traits

[lcnr]

extracted from #131813

trait Super {
    fn super_method(&self) {}
}
trait WithoutAuto: Super {}
trait WithAutoAfter: Super + Send {}
trait WithAutoBefore: Send + Super {}

impl Super for () {}
impl WithoutAuto for () {}
impl WithAutoAfter for () {}
impl WithAutoBefore for () {}

pub fn main() {
    drop::<&dyn WithoutAuto>(&());
    drop::<&dyn WithAutoAfter>(&());
    drop::<&dyn WithAutoBefore>(&());
}

results in the following assembly:

example::main::h981403045b91afa3:
        push    rax
        mov     edi, 1
        lea     rsi, [rip + .L__unnamed_1]
        call    qword ptr [rip + core::mem::drop::hc9ab6a5c0d43ca28@GOTPCREL]
        mov     edi, 1
        lea     rsi, [rip + .L__unnamed_2]
        call    qword ptr [rip + core::mem::drop::ha5195050a20993aa@GOTPCREL]
        mov     edi, 1
        lea     rsi, [rip + .L__unnamed_1]
        call    qword ptr [rip + core::mem::drop::h1632a30342c576a4@GOTPCREL]
        pop     rax
        ret

.L__unnamed_1:
        .asciz  "\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\001\000\000\000\000\000\000"
        .quad   example::Super::super_method::h936b862074aa3dae

.L__unnamed_3:
        .asciz  "\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\001\000\000\000\000\000\000"

.L__unnamed_2:
        .asciz  "\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\000\001\000\000\000\000\000\000"
        .quad   example::Super::super_method::h936b862074aa3dae
        .quad   .L__unnamed_3

https://rust.godbolt.org/z/Yef9xdjfP

This looks like the vtable for WithAutoAfter - but not for WithAutoBefore - includes a slot for Send. Given that Send is an auto-trait this seems very unexpected to me

[WaffleLapkin]

I think this is a dupe of #114942. We should never emit a vptr for traits without own vtable entries.