Provide async_read_until that uses symmetric transfer

A coroutine based implementation of `asio::async_read_until` should be more efficient than the Asio version since it can use symmetric-transfer to avoid a reschedule operation when the buffer already contain the separator.

It would be nice if you could provide this as an example. It would also be an opportunity to show users what kind of improvement they get in comparison to Asio. I would also include this in the benchmark session.

Start by locating the existing asio::async_read_until usage or examples and the benchmark session in the repository. Compare the coroutine-based implementation with Asio, focusing on the symmetric-transfer path when the buffer already contains the separator. Done means an example is provided and the improvement is included in the benchmark session.

Written by the indexing model from the issue text.