If there's a need, we can have separate machine-independent linux-errno,
openbsd-errno, cygwin-errno, etc. (as listed in
https://misc.lassi.io/2020/errno-conflicts.text) But I think the mapping
with the simplest name, "errno", should be for the current machine.
Otherwise the simplest use cases have to find out what OS is underneath
and do a case statement based on that. It all gets very complicated if
you just want to wrap some FFI code.