BytesMut::advance no longer advances cursor

[camshaft]

In #698, the behavior of BytesMut::advance was changed to clear the buffer instead of reduce capacity when the advanced count matched remaining. In s2n-quic, we rely on advance actually reducing remaining and have several data structures and trait impls that rely on this behavior:

• buffer::reassembler::Slot
• impl buffer::reader::Storage for BytesMut

In its current state, there's no longer an efficient way to actually reduce capacity of BytesMut, since the only other option is to call split_to and discard the result, which isn't great since you churn on reference counts, which is kind of the whole point of advance, as indicated by the #[must_use] attribute: consider BytesMut::advance if you don't need the other half.

My vote would be to revert this change.

[seanmonstar]

Is there a simple test we could add with a comment to make sure we don't change this behavior again?

[camshaft]

Is there a simple test we could add with a comment to make sure we don't change this behavior again?

I can open a PR tomorrow if no one beats me to it

[Darksonn]

Yes please, that would be great.

[camshaft]

Opened #728