Big endian vs little endian is a pain. I had to go through two conversions to serialize an `std::net::Ipv4Addr` to a big-endian [u8; 4] - the middle conversion was a u32 of native endianness.

The most surprising thing is that the compiler optimizes this down to nothing. Maybe the IP addresses are already stored internally as big endian? Would make sense to store them in the same order they’re commonly used.