The NATS codec has been broken due to premature optimizations that actually didn't work. I'm thinking of another way to do so.

The problem being that the old way of doing things was just reading the buffer by borrowing and copying data all around, creating unreasonable slowness on some cases.

Right now it needs to be rewritten IMO since we actually need to make sure we have a correct command to parse before splitting off the buffer and consuming that split in the parser.