Should we divide `aarch64-unknown-none` into two targets, one with floating point enabled and on without? CC @rust-embedded/cortex-a