io: add T: Unpin bound to ReadHalf::unsplit

[taiki-e]

Motivation

Fixes #5372

Solution

Adds T: Unpin bound to ReadHalf::unsplit.
Similar functions in futures also have the same bound, I believe this should be a reasonable fix here.

https://docs.rs/futures/latest/futures/io/struct.ReadHalf.html#impl-ReadHalf%3CT%3E
https://docs.rs/futures/latest/futures/stream/struct.SplitStream.html#impl-SplitStream%3CS%3E

[taiki-e]

Hmm, it seems that the change that should be backported must be sent the patch to the old LTS version first?

[notgull]

Isn't this a breaking change?

[taiki-e]

A braking change to fix soundness issues is allowed (see also rfc1122).

It depends on the number of crates affected by the breakage and whether there are other (good) ways to fix it, but I don't know of any way to fix this other than the "(auto-ref) specialization + run-time check + run-time panic" (which is not very good).

[carllerche]

Thanks, the patch looks good, but we need to apply this to the oldest LTS branch first, release, and forward port.

This is similar to #5336 and the "merge changes" https://github.com/tokio-rs/tokio/pulls?q=is%3Apr+author%3Acarllerche+is%3Aclosed

This lets us maintain the history.

I can work on it Tuesday, but if you are able to start the patching, I would appreciate it! For the patches to the LTS branches you can combine the fix + the "release" changes. like #5336 does.

[carllerche]

Re: "breaking change" I agree w/ @taiki-e. This is to fix a soundness issue, and we are taking the least intrusive approach. It shouldn't break anyone's code. Any code that does break is unsound and should be fixed.

[taiki-e]

Changed base branch to tokio-1.18.x and added "prepare release" commit.

I will leave it to carllerche for other branches. (I don't think I'm awake at the time this is merged, because it's night in Japan.)