Misc cleanups

[bjorn3]

Last PR for today. This time I didn't base it on any non-merged PR.

[ltratt]

Other than "feature gates are undesirable" (which I agree with), is there any motivation for 9146db9? The reason I went with !Sync/!Send is that it's a well-understood way of documenting that a struct isn't send/sync. Although one gets the same type safety from a phantom type, it doesn't quite have the same consistency of documentation. I can be convinced on this, as I tried both designs, and wasn't 100% convinced I chose the best.

[bjorn3]

rust-lang/rust#45198 used PhantomData<*mut Fn()> to prevent all OIBITs. That is stronger than we need here, but I believe PhantomData<...> is generally used over impl !... for ... in the standard library.

[ltratt]

OK, I can see some instances of this idiom in the standard library (not many, but a few), so I'm semi-swayed. Let's use their precedent and call the attribute either marker or (probably better) _dont_send_or_sync_me. I think we can delete the doc test too. If Rust makes raw pointers Send or Sync, this entire code base will collapse in on itself ;)

[bjorn3]

Renamed and dropped doc comment.

[ltratt]

Please squash.

[bjorn3]

Splat

[ltratt]

bors r+

[bors]

Build failed:

• buildbot/ykrustc-buildbot-build-script-hwt

[bjorn3]

I dropped the first commit. Locally it showed an unused import warning and compiled fine with removing the import, but on CI it is necessary.

[ltratt]

bors r+

[bors]

Build succeeded:

• buildbot/ykrustc-buildbot-build-script-hwt