[Tracking] Potential performance improvements

[lnicola]

In my quick tests, there is some room for improvement in the performance of this crate. We discussed some ideas in #23, mainly:

• increasing the read buffer size or making it configurable
• using BytesMut to reduce allocations
• introducing a small amount buffering so that one chunk can be read from disk while another is being sent over the network

I'm filing this as somewhat of a tracking issue, because I don't think it's worth trying to work on it right now. The buffer size change is partially blocked because tokio limits the read size to 16 KB (tokio-rs/tokio#1976). And in my tests I couldn't get consistent performance, likely because of extra context switching in tokio (tokio-rs/tokio#1844).