Proposal: Faster netstack

I propose that we rewrite the Redox netstack with the following features:

• designed for high socket counts/many connections and maximum throughput
• zero-copy data transfers between drivers and netstack where possible (io_uring?)
• multithreaded processing for checksums and transfer to socket owner, perhaps tuned to available core count
• IP only in the netstack daemon, TCP and higher protocols in relibc of the socket owner
• See also QUIC
• Consider also how to implement BSD sockets and sending stateful fds, perhaps in a separate daemon