netlink-proto: refactor

[little-dude]

This is just a couple notes to myself regarding some netlink-proto simplifications I've been contemplating for a while:

• have a distinct codec for the audit protocol, instead of using feature flags
• don't use the tokio codec trait? They force us to copy the data in intermediate buffers, which most of the time we don't need (unless we read multiple datagrams at once, like with the audit protocol).
• have NetlinkFramed return a Result instead of an Option
• rename NetlinkFramed into Transport maybe?
• don't use unbounded channels? But then we have to handle errors when they get full. Not sure how to do that

Also:

• https://github.com/little-dude/netlink/issues/138
• https://github.com/little-dude/netlink/issues/139

[little-dude]

Also noticed that we need to upgrade to newer tokio-utils and bytes versions.