Possibly refactor AsyncBufRead

[Darksonn]

In the spirit of this RFC, we should consider refactoring AsyncBufRead in a similar way to allow unlimited peeking.

It is worth noting that this removes some types' ability to implement the trait without complicating the types. This is the case for e.g. StreamReader. That said, as is, the trait is more or less only useful for read_line, and other single-byte-separated splits.

[seanmonstar]

For 0.3, I'd want to propose just removing the trait, and figuring it out when not under a time crunch.

[seanmonstar]

I'm still personally in favor or removing the trait and suggesting people use LinesCodec if they want that feature. However, if keeping it, here's a proposal to allow us to punt design:

Seal AsyncBufRead

We could seal the trait, similar to ToSocketAddrs. That means all the methods would be "private", and it's only usefulness is possible as bounds, and the extension methods like lines().

[carllerche]

I believe that based on #2896, we will need to move AsyncBufRead -> tokio-util? It probably would be nice to include some note in the Tokio docs pointing to tokio-util.

[carllerche]

We are going to punt on this for now.