A team at Google is working on an embedded project in which they've modified core::fmt to remove floating point operations. This has resulted in the Display impls for byteorder::F32 and F64 to fail to build.

It's common for kernel or embedded projects to build without support for floating point operations. We should make it so that the F32 and F64 types (and any other floating point types or operations I haven't thought of) can be disabled.

In 0.9, we can add an on-by-default feature for floating point operations which users can disable if they want. We might want to do that for 0.8 as well, but that would risk causing breakage for customers who depend on zerocopy 0.8 with default features disabled. Another approach would be to add a --cfg for 0.8 and then use an on-by-default feature in 0.9.