cast::transmute shows up in profiles

[jdm]

Even though cast::transmute is inlined, apparently the rusti::transmute intrinsic is not. Perf shows this annotated view:

       │    0000000000088dd0 <cast::transmute_9885::_c97de6de549dd4c9::_07pre>:
  4.56 │      cmp    %fs:0x70,%rsp
       │    ↓ ja     25
       │      movabs $0x28,%r10
       │      movabs $0x0,%r11
       │    → callq  __morestack
       │    ← retq
 14.10 │25:   push   %rbp
  0.43 │      mov    %rsp,%rbp
  2.17 │      sub    $0x20,%rsp
  2.82 │      mov    %rdx,-0x20(%rbp)
  1.95 │    ↓ jmp    3d
       │33:   mov    -0x8(%rbp),%rax
 24.30 │      add    $0x20,%rsp
  0.22 │      pop    %rbp
  0.22 │    ← retq
  0.22 │3d:   mov    -0x20(%rbp),%rax
  8.68 │      mov    %rax,-0x10(%rbp)
  4.12 │      mov    -0x10(%rbp),%rax
 10.63 │      mov    %rax,-0x18(%rbp)
  1.30 │      mov    -0x18(%rbp),%rdx
 15.40 │    → callq  cast::rusti::transmute_9888::_a9e57e9b33a18c23::_07pre
       │      mov    %rax,-0x8(%rbp)
  8.89 │    ↑ jmp    33

This is showing up in servo profiles.

[Aatch]

Our codegen for the transmute intrinsic is wrong at the moment, use one and check the lint output. This means that LLVM often refuses to inline it. That is why it shows up in profiles.

[dotdash]

This should be done. Intrinsics codegen is better now, and they are always inlined now.

[alexcrichton]

Would this be a good candidate for a codegen test with @graydon's new framework?